Three-dimensional base calling in next-generation sequencing analysis

Abstract

Disclosed herein are sequencing systems and sequencing methods for training a neural network and for performing sequencing analysis with the trained neural network after acquiring a flow cell image using the sequencing system. The sequencing systems disclosed herein can include a field programmable gate array (FPGA), an artificial intelligence (AI) chip, or a combination thereof.

Description

[0001] Cross-references

[0002] This application claims the benefits of U.S. Provisional Application No. 63 / 549,327, filed February 2, 2024; U.S. Provisional Application No. 63 / 549,333, filed February 2, 2024; U.S. Provisional Application No. 63 / 570,038, filed March 26, 2024; U.S. Provisional Application No. 63 / 661,332, filed June 18, 2024; U.S. Provisional Application No. 63 / 724,712, filed November 25, 2024; and U.S. Provisional Application No. 63 / 736,743, filed December 20, 2024, each of which is incorporated herein by reference in its entirety. Technical Field

[0003] Embodiments of this disclosure generally relate to image processing and base interpretation in sequencing data analysis, and specifically to three-dimensional (3D) images of in situ samples. Background Technology

[0004] In next-generation sequencing (NGS) or NGS-like applications (such as synthesis sequencing, binding sequencing, or affinity sequencing), a new strand is synthesized one nucleotide base at a time to identify the sequence of a target nucleic acid. In each sequencing cycle, one base is attached to any given strand. In the imaging step of each cycle, an image is recorded. Base interpretation algorithms are applied to the image to "read" the continuous signals from each cluster or polymerase clone and convert the optical signals into recognition of the nucleotide base sequence added to each DNA fragment. Traditional sequencing data analysis relies on two-dimensional (2D) flow cell images. When sequencing analysis is performed on in situ samples such as cells or tissues, the sample has thickness along the z-direction perpendicular to the image plane. Therefore, flow cell images at selected z-levels can include signals from defocused polymerase clones located at adjacent z-levels and other undesirable signals, such as signals from the cell membrane. There is a need for rapid and accurate three-dimensional (3D) flow cell image processing and base interpretation to ensure reliable base interpretation and sequencing analysis of 3D samples such as cells and tissues. Summary of the Invention

[0005] This document provides embodiments of systems, apparatuses, methods, and / or computer programs, and / or combinations and sub-combinations thereof, which enable rapid and accurate flow cell image processing for reliable and accurate base interpretation of samples such as in situ cells or tissues. Flow cell images may originate from different sequencing cycles and / or different channels.

[0006] As a specific application of this, this document discloses embodiments of methods, systems, and media for image processing of flow cell images of 3D volumetric samples (e.g., cells or tissues), enabling accurate base interpretation based on image intensity, cluster or polymerase clone location, and / or size. The image processing method described herein reverses the imaging process of the optical system and effectively improves the full width at half maximum (FWHM) of the optical system. Therefore, the image processing method disclosed herein can advantageously increase the detectable density of polymerase clones or clusters in 3D samples or conventional 2D samples. The method described herein can advantageously reduce the effects of color mixing of polymerase clones that may be caused by neighboring polymerase clones in the 2D or 3D dimension by computationally improving the spatial resolution of the flow cell image.

[0007] In some embodiments, a neural network (e.g., a convolutional neural network) is used to generate a high-resolution z-stacking of a 3D sample flow cell image from a low-resolution z-stacking obtained from a sequencing system, and subsequent primary analyses can be performed based on the high-resolution flow cell image instead of the low-resolution flow cell image. In some embodiments, a neural network (e.g., a convolutional neural network) is used in the image processing of the high-resolution z-stacking of the sample's flow cell image to generate base interpretations.

[0008] Embodiments of these aspects include corresponding computer systems, devices, and computer program products recorded on computer storage devices, configured individually or in combination to perform the operations of the method. For a computer system configured or to be configured to perform the operations, software, firmware, hardware, or combinations thereof are installed thereon to cause the computer system to perform the operations or actions in operation. For a computer program product configured or to be configured to perform the operations or actions, the computer program product includes instructions that, when executed by a hardware processor, cause the hardware processor to perform the operations or actions.

[0009] Further embodiments, features, and advantages of this disclosure, as well as the structure and operation of various embodiments of this disclosure, are described in detail below with reference to the accompanying drawings. Attached Figure Description

[0010] The accompanying drawings, which are incorporated herein and form a part of this specification, illustrate embodiments of the present disclosure and, together with this specification, further serve to explain the principles of the present disclosure and enable those skilled in the art to make and use these embodiments.

[0011] Figure 1 A block diagram of a sequencing system according to some embodiments for performing sequencing, flow cell image processing and / or primary analysis operations (including base interpretation using flow cell images) is illustrated.

[0012] Figures 2A to 2CAn exemplary simulated flow cell image is shown according to some embodiments ( Figure 2A Images of in situ cell samples) and predictions and corresponding data using the systems and methods described in this paper. Figure 2A Two different images in the middle ( Figures 2B to 2C The predicted images are at different z-levels.

[0013] Figures 2D to 2E Exemplary simulated flow cell images from a reference set are shown. According to some embodiments, images with a first resolution (…) are generated using the methods described herein. Figure 2D ) and second resolution ( Figure 2E The simulated flow cell image.

[0014] Figures 3A to 3D Two exemplary flow cell images with multiple cells at two different z-levels are shown. Figure 3A and Figure 3D ), and according to some embodiments, using the systems and methods described herein from Figure 3A The image generated from the image at two different z-levels ( Figures 3B to 3C ).

[0015] Figures 3E to 3F This demonstrates that the method presented in this paper improves the detection rate of each cell of the target within the same imaging region compared to non-AI-based methods. Figure 3E And the false positive rate is low. Figure 3F In this case, the target is a polymerase clone or cluster within the cell.

[0016] Figure 3G The paper demonstrates that the neural network proposed in this paper improves the detection of targets in simulated flow cell images of samples, achieving higher R-values ​​than conventional methods. 2 value.

[0017] Figure 4 A block diagram is illustrated of a computer system, according to some embodiments, for performing image processing, sequencing analysis, training neural networks, predicting base interpretation results, image intensity, high-resolution images, and / or classification and / or base interpretation using a pre-trained neural network.

[0018] Figure 5A This is a flowchart of an exemplary method for predicting a 3D flow cell image of a sequencing sample and performing base interpretation using that 3D flow cell image, according to some embodiments.

[0019] Figure 5B This is a flowchart of an exemplary method for training a neural network that can be used to predict high-resolution flow cell images of sequencing samples, according to some embodiments.

[0020] Figure 5C The diagram illustrates an exemplary embodiment of a first reconfigurable logic device, an integrated circuit, and its connection to a processor of a sequencing system.

[0021] Figure 5D As an illustration, it shows an exemplary embodiment in which a first reconfigurable logic device and integrated circuit are used in parallel with an ongoing sequencing run within a predetermined time window.

[0022] Figure 5E The flowchart illustrates an exemplary method for training a neural network according to some embodiments, thereby generating a pre-trained neural network that can be used to predict higher resolution flow cell images, base interpretation results, intensity, and / or classification of sequencing samples.

[0023] Figure 5F A scatter plot of an exemplary embodiment of generating reference intensity from a high-resolution training flow cell image is shown.

[0024] Figure 6 A schematic diagram illustrating an exemplary embodiment of a locking probe.

[0025] Figure 7 To illustrate a workflow for generating circularized locked probes in cells, the workflow includes generating a first cDNA and a second cDNA from a first target RNA molecule and a second target RNA molecule, respectively, and hybridizing the first locked probe and the second locked probe with the first cDNA and the second cDNA molecule, respectively, to generate the first circularized locked probe and the second circularized locked probe, respectively.

[0026] Figure 8 To illustrate the intracellular rolling circle and sequencing workflow, this workflow includes generating a first and a second polyplast by rolling circle amplification (respectively) using a first covalently closed circular molecule and a second covalently closed circular molecule. The first and second polyps undergo a sequencing workflow using universal sequencing primers, sequencing polymerase, and various nucleotide reagents.

[0027] Figure 9 This is a schematic diagram illustrating an exemplary workflow for sequencing multiply generated inside a cell.

[0028] Figure 10 This is a schematic diagram illustrating an exemplary workflow for sequencing multiply generated inside a cell.

[0029] Figure 11 This is a schematic diagram illustrating an exemplary workflow for sequencing multiply generated inside a cell.

[0030] Figure 12This is a schematic diagram illustrating an exemplary workflow for sequencing multiply generated inside a cell.

[0031] Figure 13 To illustrate a workflow for generating circularized locked probes, the workflow includes generating a first cDNA and a second cDNA from a first target RNA molecule and a second target RNA molecule, respectively, and hybridizing the first locked probe and the second locked probe with the first cDNA and the second cDNA molecule, respectively, to generate the first circularized locked probe and the second circularized locked probe, respectively.

[0032] Figure 14 To illustrate the rolling circle and sequencing workflow, the workflow includes generating a first multiply and a second multiply by rolling circle amplification using a first covalently closed circular molecule and a second covalently closed circular molecule, respectively.

[0033] Figure 15 This is a schematic diagram of an exemplary low-binding support, which includes a glass substrate and alternating hydrophilic coatings covalently or non-covalently adhered to the glass, and further includes chemically reactive functional groups that act as attachment sites for oligonucleotide primers (e.g., capturing oligonucleotides).

[0034] Figure 16 Schematic diagrams of various exemplary configurations of multivalent molecules. Left (Class I): Schematic diagram of a multivalent molecule with a "starburst" or "helter-skelter" configuration. Middle (Class II): Schematic diagram of a multivalent molecule with a dendritic macromolecular configuration. Right (Class III): Schematic diagram of multiple multivalent molecules formed by the reaction of streptavidin with a 4-arm or 8-arm PEG-NHS containing biotin and dNTPs. Nucleotide units are designated as 'N', biotin is designated as 'B', and streptavidin is designated as 'SA'.

[0035] Figure 17 A schematic diagram of an exemplary multivalent molecule comprising a universal core attached to multiple nucleotide arms.

[0036] Figure 18 This is a schematic diagram of an exemplary multivalent molecule comprising a dendritic macromolecular core attached to multiple nucleotide arms.

[0037] Figure 19 A schematic diagram of an exemplary multivalent molecule is shown, comprising a core attached to multiple nucleotide arms, wherein the nucleotide arms contain biotin, spacers, linkers, and nucleotide units.

[0038] Figure 20 This is a schematic diagram of an exemplary nucleotide arm that includes a core attachment portion, spacers, linkers, and nucleotide units.

[0039] Figure 21 The chemical structures of exemplary spacers (top) and various exemplary connectors, including 11-atom connectors, 16-atom connectors, 23-atom connectors and N3 connectors (bottom), are shown.

[0040] Figure 22 The chemical structures of various exemplary connectors (including connectors 1 to 9) are shown.

[0041] Figure 23A The chemical structures of various exemplary linkers that connect / attach to nucleotide units are shown.

[0042] Figure 23B The chemical structures of various exemplary linkers that connect / attach to nucleotide units are shown.

[0043] Figure 23C The chemical structures of various exemplary linkers that connect / attach to nucleotide units are shown.

[0044] Figure 23D The chemical structures of various exemplary linkers that connect / attach to nucleotide units are shown.

[0045] Figure 24 The chemical structure of an exemplary biotinylated nucleotide arm is shown.

[0046] Figure 25 A schematic diagram of guanine tetrads (e.g., G-tetrads).

[0047] Figure 26 This is a schematic diagram of an exemplary intramolecular G-quadruplex structure.

[0048] Figure 27 An exemplary support according to some embodiments is shown, which has multiple blocks for fixing 2D or 3D samples (including cell samples) thereon for sequencing.

[0049] Figure 28 A flowchart is shown of an exemplary method for predicting base interpretation results of flow cell images (e.g., of in-situ samples) using neural networks disclosed herein, according to some embodiments.

[0050] Figure 29 A flowchart illustrating an exemplary method for training a neural network that can be used to predict base interpretation results or high-resolution flow cell images, according to some embodiments, is shown.

[0051] Figures 30A to 30B The flow cell image is shown. Figure 30A(This refers to) and its high-resolution image predicted by a neural network pre-trained using reference base interpretation results. In this case, the base interpretation results are determined from the high-resolution image using an algorithm that is not based on a neural network.

[0052] Figure 31 A block diagram illustrating an exemplary method for training a neural network, and an exemplary method for using such a pre-trained neural network to predict high-resolution flow cell images and / or predict base interpretation results.

[0053] In the accompanying drawings, similar reference numerals typically denote the same or similar elements. Additionally, the leftmost numeral in the reference numerals usually identifies the drawing in which the reference numeral first appears. Detailed Implementation

[0054] This document provides embodiments of systems, apparatuses, methods, and / or computer program products, and / or combinations and sub-combinations thereof, which implement image processing of flow cell images (e.g., flow cell images obtained from in-situ samples or conventional 2D samples during a sequencing run) to: 1) generate images with improved spatial resolution and improved detectable polymerase clone or cluster density, perform base interpretation using flow cell images with such improved spatial resolution, and the generated images can be used for subsequent sequencing analyses, including but not limited to base interpretation; or 2) predict the strength, base interpretation result, or classification of polymerase clones or clusters. The techniques described herein can be used while a sequencing run is still in progress to improve the efficiency of sequencing and sequencing analysis, reduce the data storage required during sequencing and sequencing analysis, and improve the accuracy and reliability of sequencing analysis. The techniques described herein can be used with flow cell images obtained using various imaging and / or sequencing techniques with volumetric 3D samples and / or conventional 2D samples, and / or using various sequencing systems (e.g., next-generation sequencing (NGS) systems). The techniques disclosed herein can be used for base interpretation in NGS, and NGS flow cell images will be used as primary examples of applications for which these techniques are described herein. However, such image analysis techniques may also be useful in other applications using point detection and / or CCD imaging.

[0055] Traditional flow cell images can show clusters or polymerase clones from 2D samples, and base interpretation can be performed using their corresponding image intensities. Existing base interpretation algorithms may fail to distinguish polymerase clones or clusters from background noise or other signal points of similar shape or size. For accurate and reliable base interpretation, there is a need to identify the location of polymerase clones or clusters, preferably with improved spatial resolution compared to the flow cell image, especially for samples with increased polymerase clone or cluster density compared to conventional 2D samples. There is also a need to generate accurate and reliable intensity values ​​for polymerase clones or clusters so that they can be used for accurate and reliable base interpretation. Furthermore, it is necessary to use information included in the flow cell image (such as polymerase clone location, shape, intensity, etc.) to accurately predict base interpretations.

[0056] The techniques described herein can be used to process flow cell images (e.g., 2D or 3D) to generate accurate and reliable image intensities of polymerase clones or clusters with improved spatial resolution, and thus improved maximum polymerase clone or cluster density detected in a sample for accurate and reliable sequencing analysis. The techniques disclosed herein can advantageously reverse the imaging process of an optical system and effectively improve the full width at half maximum (FWHM) of the imager, so that the density of polymerase clone sites is not limited by the optical design of the sequencing system. Therefore, the disclosed techniques can advantageously increase the detected polymerase clone density, for example, by a factor of 2, 4, 8, 16, 27, 40, 50, 100, or higher compared to densities detectable using conventional optical systems and image processing methods. In some embodiments, the disclosed techniques can advantageously increase the spatial resolution of flow cell images in each of one or more spatial dimensions by a factor of 2, 4, 8, 16, 27, 40, 50, 100, or higher compared to flow cell images acquired using conventional optical systems and / or image processing methods. The method presented in this paper can also advantageously reduce the effects of color mixing of polymerase clones that may be caused by neighboring polymerase clones or clusters by computationally increasing the spatial resolution of the flow cell image.

[0057] In situ samples, such as cells or tissues, may not remain in focus in a single 2D image due to thickness along the axial or z-direction. Multiple 2D flow cell images can be stacked in z-columns to cover clusters or polymerase clones at different z-levels, for example, in 3D cell samples. Interference can occur in the z-column of flow cell images, such as out-of-focus polymerase clones and background signals from cellular components. For example, a polymerase clone located at a first z-level may appear in a first flow cell image at that first z-level, and it may also generate a signal patch in a second 2D flow cell image acquired at its adjacent z-level, which is out of focus. This signal patch may interfere with the intensity of polymerase clones at or near the same xy location in the second flow cell image, thus reducing the accuracy and reliability of base interpretation. As another example, color mixing from neighboring polymerase clones may interfere with the polymerase clone intensity or density that can be detected for subsequent base interpretation. There is a need for identifying polymerase clone or cluster locations for accurate and reliable 3D base interpretation. There is also a need for generating accurate and reliable intensity values ​​from polymerase clones or clusters in 3D volume samples, so that they can be used for accurate and reliable base interpretation.

[0058] The techniques disclosed herein advantageously train neural networks to efficiently and accurately predict the location of polymerase clones or clusters in a sample. These techniques also advantageously train neural networks to efficiently and accurately predict the high-resolution intensity, base interpretation results, and / or classification of polymerase clones or clusters in a sample. The samples described herein are not limited to 3D samples, such as in situ cells and / or tissues. The samples described herein may also include conventional 2D samples.

[0059] The techniques disclosed herein can advantageously utilize reconfigurable logic devices (e.g., FPGAs) and other integrated circuits (e.g., AI chips or neural processing units (NPUs)) to: 1) predict high-resolution polymerase clone or cluster locations based on low-resolution flow cell images; or 2) predict the intensity, base interpretation, and / or classification of polymerase clones or clusters in a sample at high resolution. When compared with sequencing analyses using existing sequencing systems, utilizing reconfigurable logic devices (e.g., FPGAs) and other integrated circuits (e.g., AI chips or neural processing units (NPUs)) on sequencing systems can advantageously reduce computation time, energy consumption, improve sequencing analysis efficiency, reduce required data storage space, and lower sequencing system costs in the analysis of flow cell images.

[0060] The technique disclosed herein is based on a loss function determined by comparison with a reference base interpretation result (as a baseline truth). The trained neural network can then be used to accurately and reliably predict high-resolution flow cell images after image processing, based on flow cell images acquired from samples. The technique disclosed herein advantageously allows for mismatches between training and prediction outputs. For example, the neural network can be trained by generating a training base interpretation result as the training output and comparing this output with a reference base interpretation result (as a baseline truth). The trained neural network can then be used to predict either high-resolution flow cell images or base interpretation results. Such mismatches between training and prediction outputs advantageously allow for the consideration of the reference base interpretation result in the training parameters of the neural network, and for predicting higher-resolution, higher-quality versions of flow cell images, which can be used to improve the accuracy and reliability of base interpretation. This type of training and prediction advantageously enables the use of simplified neural networks requiring less computation, reducing computation time, power consumption, and the number of predictions made.

[0061] The samples described herein are not limited to 3D samples, such as in situ cells and / or tissues. The samples described herein may also include conventional 2D samples. The techniques disclosed herein can advantageously utilize reconfigurable logic devices (e.g., FPGAs) and other integrated circuits (e.g., AI chips or neural processing units (NPUs)) to perform one or more operations in training and / or prediction.

[0062] In DNA sequencing, identifying the center of clusters or polymerase clones is sometimes considered part of the primary analysis. Primary analysis may include some or all of the operations and / or steps required to perform base interpretation and calculate the quality score of the base interpretation. Primary analysis may involve forming a template image for at least a portion of the flow cell. The template image may include the estimated location of all detected clusters or polymerase clones in a common coordinate system. The template image may include a 2D or 3D polymerase clone map. The template image is generated by identifying the locations of clusters or polymerase clones in all images during the first cycle or the first few cycles of the sequencing method. The generation of the template image may require sufficient spatial resolution to distinguish polymerase clones from background features, neighboring polymerase clones, and / or out-of-focus duplicate polymerase clones.

[0063] sequencing system

[0064] Figure 1A block diagram of a computer-implemented system 100 according to one or more embodiments disclosed herein is illustrated. System 100 has a sequencing system 110 including a flow cell 112, a sequencer 114, an imager 116, a data storage 122, and a user interface 124. Sequencing system 110 is connectable to a cloud 130. Sequencing system 110 may include one or more of a dedicated processor 118, a first reconfigurable logic device (e.g., a field-programmable gate array (FPGA)) 120, and a computing system 126.

[0065] In some embodiments, flow cell 112 is configured to capture DNA fragments and form DNA sequences for base interpretation on the flow cell. Flow cell 112 may include the support disclosed herein. The support may be a solid support. As disclosed herein, the support may include a surface coating thereon. The surface coating may be a polymer coating as disclosed herein.

[0066] Flow cell 112 may include multiple blocks or imaging regions thereon, and each block may be divided into a sub-block grid. Each sub-block may include multiple clusters or polymerase clones fixed thereon. As a non-limiting example, the flow cell may have 424 blocks, and each block may be divided into a 6 x 9 grid, thus having 54 sub-blocks. Flow cell images as disclosed herein may be images of signals including multiple clusters or polymerase clones. Flow cell images may include one or more signal blocks or one or more signal sub-blocks. In some embodiments, a flow cell image may be an image including all blocks and approximately all signals thereon. Flow cell images may be acquired from channels using imager 116 during an imaging or sequencing cycle. In some embodiments, each block may include millions of polymerase clones or clusters. As a non-limiting example, a block may include approximately 1 million to 10 million clusters or polymerase clones. Each polymerase clone may be a collection of many copies of DNA fragments.

[0067] Based on a sample fixed to a support (e.g., a flow cell), flow cell images can be acquired using imager 116 at one or more z-levels along a z-axis orthogonal to the image plane of the flow cell image. Specifically, for three-dimensional samples, such as cells, tissues, or other... In situThe flow cell image may include multiple z-levels (i.e., z-layers) to cover the entire sample in 3D. The z-axis may extend from the objective lens of the imager 116 disclosed herein to a support, such as the flow cell 112. The z-axis may be orthogonal to the image plane of the flow cell image. Each z-level of the flow cell image may be separated from adjacent z-levels by a predetermined distance, for example, ranging from about 0.1 μm to about 15 μm or from 0.02 μm to 10 μm. Each z-level of the flow cell image may be separated from adjacent z-levels by a distance ranging from 0.5 μm to 10 μm, 0.01 μm to 5 μm, or 0.1 μm to 15 μm. At each z-level, the flow cell image may be acquired from one or more sequencing cycles and / or one or more channels. Each flow cell image may include at least a portion of one or more blocks or sub-blocks of the flow cell within its field of view. Figure 27 A portion of a flow cell 2712 with multiple blocks 2710 is shown. The image plane is defined by the x-axis and y-axis. The z-direction (i.e., the z-axis) is orthogonal to the xy-plane. Although... Figure 27 As shown, the flow cell image, sample, and z-axis are described in Cartesian coordinates, but any other coordinate system can be used to define the spatial positions and relationships in this paper. Other coordinate systems may include, but are not limited to, polar, cylindrical, or spherical coordinates.

[0068] The sequencer 114 can be configured to flow a mixture of nucleotides onto a flow cell 112, remove a nucleotide cleavage inhibitor between flow steps, and perform other steps for forming a DNA sequence on the flow cell 112. The nucleotides may have attached fluorescent elements that emit light or energy at wavelengths indicating the type of nucleotide. Each type of fluorescent element may correspond to a specific nucleotide base (e.g., A, G, C, T). The fluorescent elements may emit light at visible wavelengths. In some embodiments, the sequencer 114 and flow cell 112 can be configured to perform various sequencing methods disclosed herein, such as affinity sequencing.

[0069] For example, each nucleotide base can be assigned a color. Different types of nucleotides can have different colors. For example, adenine (A) can be red, cytosine (C) can be blue, guanine (G) can be green, and thymine (T) can be yellow. The color or wavelength of the fluorescent element can be selected for each nucleotide, so that nucleotides can be distinguished from each other based on the wavelength of light emitted by the fluorescent element.

[0070] Imager 116 may be configured to capture an image of flow cell 112 after each flow step. In some embodiments, imager 116 includes a camera, such as a CMOS or CCD camera, configured to capture digital images. The camera may be configured to capture images of the wavelengths of fluorescent elements that bind to nucleotides. An image of a sample immobilized on at least a portion of the flow cell, acquired by the imager, may be referred to as a flow cell image.

[0071] In some embodiments, imager 116 may include one or more optical systems disclosed herein. The optical systems may be configured to capture optical signals from the flow cell and generate a corresponding flow cell image. The flow cell image can then be used for base interpretation.

[0072] In one embodiment, images of the flow cell can be captured in groups, wherein each image in the group is taken at a wavelength or spectrum that matches or includes only one of the fluorescent elements. In another embodiment, the images can be captured as a single image capturing all wavelengths of the fluorescent elements.

[0073] The resolution of imager 116 controls the level of detail in the flow cell image, including pixel size. This resolution is critical in existing systems because it controls the accuracy of the point-finding algorithm in identifying polymerase clones or cluster centers. In some embodiments, the image resolution of the flow cell images disclosed herein can be from about 10 nanometers (nm) to hundreds of nm or higher. In some embodiments, the image resolution of the flow cell image can range from 0.1 nm to 1000 nm. In some embodiments, the image resolution of the flow cell image can range from 1 nm to 500 nm. In some embodiments, the image resolution of the flow cell image can range from 5 nm to 300 nm. One way to improve the accuracy of polymerase clone or cluster finding is to improve the resolution of imager 116 or to improve the processing performed on the images acquired by imager 116. Detection of polymerase clones or cluster centers in pixels other than those detected by the point-finding algorithm can be performed. These methods can allow for improved accuracy in polymerase clone or cluster center detection without increasing the resolution of imager 116. The resolution of imager 116 can even outperform existing systems with comparable performance, which can reduce the cost of sequencing system 110.

[0074] The image quality of the flow cell image controls the quality of base interpretation. One way to increase the accuracy of base interpretation is to improve the imager 116 or improve the processing performed on the images acquired by the imager 116 to obtain better image quality. The methods described herein can predict high-resolution flow cell images (2, 4, or higher than the resolution of existing flow cell images in a common coordinate system), making it possible to improve the detectable polymerase clone or cluster density while reducing or eliminating interference from neighboring polymerase clones, cellular background signals, color mixing, and / or other noise in the flow cell image. Therefore, 3D base interpretation using the methods of this paper may be more accurate when compared with existing methods that do not use such high-resolution flow cell images. Such methods of this paper allow for accurate and efficient base interpretation. These methods can be advantageously executed in parallel with sequencing runs in a computer-implemented system 100 without interfering with or delaying the existing sequencing workflow of the sequencing system 110. The results of the predicted high-resolution flow cell image can be used to make base interpretations in the current sequencing cycle in the sequencing workflow. Furthermore, some or all of the operations disclosed herein can be advantageously performed by a first reconfigurable logic device (e.g., FPGA) or integrated circuit (e.g., application-specific integrated circuit (ASIC) chip, neural processing unit (NPU) or artificial intelligence (AI) chip), and data can be transferred between the CPU and the first reconfigurable logic device or integrated circuit to reduce the total operating time from methods that use only the CPU.

[0075] Sequencing system 110 can be configured to perform image processing operations or actions for flow cell images across different cycles and / or channels. The operations or actions disclosed herein can be performed by a dedicated processor 118, reconfigurable logic devices and / or integrated circuits 120, computing system 126, or combinations thereof. One or more operations or actions of methods 500, 600, 700, 2800, and 2900 disclosed herein can be performed by a dedicated processor 118, reconfigurable logic devices and / or integrated circuits 120, computing system 126, or combinations thereof. In some embodiments, which operations or actions will be performed by the dedicated processor 118, reconfigurable logic devices and / or integrated circuits 120, computing system 126, or combinations thereof can be determined based on one or more of the following: computation time for a particular operation, computational complexity in a particular operation, the need for data transfer between hardware devices, power consumption required for a particular operation, or combinations thereof. Image processing operations or actions for the flow cell image can be performed after acquiring the corresponding flow cell image but before performing base interpretation on the flow cell image.

[0076] In some embodiments, data storage 122 is used to store information used in the methods herein. This information may include the flow cell image itself or information and / or images derived from flow images captured by imager 116. DNA sequences determined from base interpretation may be stored in data storage 122. Parameters identifying the location of polymerase clones or clusters may also be stored in data storage 122. Raw and / or processed image intensities of each polymerase clone or cluster may be stored in data storage 122. Regions and / or sub-blocks corresponding to each polymerase clone or cluster may also be stored in data storage 122. Transformation matrices for each region and / or sub-block of different cycles and / or channels may also be stored in data storage 122. Cell images may be stored in data storage 122. Flow cell images, processed images, and / or filtered images may be stored in data storage. Other information or images that may facilitate 3D base interpretation of the sample may be stored in data storage.

[0077] The user interface 124 can be used by the user to operate the sequencing system or access data stored in the data storage 122 or the computing system 126.

[0078] The computing system 126 controls the general operation of the sequencing system and is coupled to the user interface 124. The computing system can also perform steps of image processing, base interpretation, and their pre- and / or post-processing operations, including but not limited to predicting high-resolution flow cell images. In some embodiments, the computing system 126 is a computer system 400, such as... Figure 4 As described in more detail below. The computing system 126 may store information about the operation of the sequencing system 110, such as configuration information, instructions for operating the sequencing system 110, or user information. The computing system 126 may be configured to transfer information between the sequencing system 110 and the cloud 130.

[0079] The computing system 126 may include one or more general-purpose computers that provide an interface for running various programs in an operating system such as Windows™ or Linux™. Such an operating system typically provides users with great flexibility. In some embodiments, the computing system 126 may include one or more processors, such as CPUs, which are configured for the development and training of artificial intelligence algorithms (e.g., neural network training), and may be used alone or in combination with reconfigurable logic devices and / or integrated circuits 120.

[0080] In some embodiments, the sequencing system may include one or more reconfigurable logic devices 120 and / or one or more other integrated circuits 120. The reconfigurable logic device 120 may include one or more FPGA devices. The integrated circuit 120 herein may be reconfigurable or non-reconfigurable, and may include an AI chip, an application-specific integrated circuit (ASIC) chip, a neural processing unit (NPU), or a combination thereof. In some embodiments, the reconfigurable logic device and / or integrated circuit 120 may be configured for artificial intelligence algorithm development and training (e.g., neural network training), and may be used alone or in combination with a CPU and / or GPU.

[0081] In some embodiments, the reconfigurable logic device and / or integrated circuit 120 includes a main unit and edge units. For example, the main unit may be an FPGA device, and the edge unit may be an ASIC or an AI chip. In some embodiments, the edge unit is an additional hardware processing module that can be separately mounted to and / or removed from the system 110. The edge unit may be configured for the development and training of artificial intelligence algorithms. The edge unit may be configured for inference or prediction using deployed AI algorithms (e.g., neural networks). The edge unit may communicate electronically with the main unit via a DMA connection, for example, data communication. The edge unit may communicate electronically with other parts of the system 100 via various connections, such as chip-to-chip interconnects. For example, the edge unit may include a neural processing unit (NPU) chip, an AI chip, or any other integrated circuit.

[0082] In some embodiments, a dedicated processor 118 may be configured to perform the operations of the methods disclosed herein. The dedicated processor 118 may include one or more reconfigurable logic devices and / or integrated circuits disclosed herein. The dedicated processor 118 may not include a general-purpose processor, but rather a custom processor with specific hardware or instructions for performing those steps. The dedicated processor runs specific software directly without an operating system. The lack of an operating system reduces overhead, but at the cost of reduced processor flexibility. The dedicated processor may use a custom programming language, which can be designed to operate more efficiently than software running on a general-purpose computer. This can increase the speed of execution steps and allow for real-time processing.

[0083] In some embodiments, the reconfigurable logic device and / or integrated circuit 120 (e.g., FPGA and / or AI chip) may be configured to perform some or all of the operations in the methods disclosed herein. The reconfigurable logic device and / or integrated circuit may be programmed as hardware capable of performing a specific task. Software steps may be translated into hardware components using a special programming language. Each software step may correspond to at least one operation or action in the methods disclosed herein. Each software step may include at least a portion of the operations or actions in the methods disclosed herein. Once the reconfigurable logic device is programmed, the hardware directly processes the digital data provided to it without running software. The reconfigurable logic device and / or integrated circuit instead uses logic gates and registers to process the digital data. Because the operating system requires no overhead, the reconfigurable logic device and / or integrated circuit typically processes data faster than a general-purpose computer. Similar to dedicated processors, this may come at the cost of flexibility. The lack of software overhead may also allow the reconfigurable logic device and / or integrated circuit to operate faster than a dedicated processor, although this will depend on the exact processing to be performed and the specific reconfigurable logic device and / or integrated circuit and dedicated processor.

[0084] A set of reconfigurable logic devices and / or integrated circuits 120 can be configured to perform these steps in parallel. In some embodiments, multiple processing engines of an FPGA can be configured to perform one or more identical image processing steps on selected regions of an image, image set, sub-block, or one or more images. Each FPGA 120 can perform its own portion of the image processing step in parallel, thereby reducing the time required to process data. This allows image processing steps to be completed in real time. For example, multiple processing engines of a first FPGA can be configured to generate polymerase clone maps of flow cell blocks. Each processing engine can be responsible for generating a portion of the polymerase clone map in parallel, for example, at different sub-blocks within the block, such as non-overlapping portions. A second FPGA can be configured to perform intensity normalization in parallel with the generation of the polymerase clone map. As another example, a number of FPGAs and integrated circuits (e.g., AI chips) can be configured to perform one or more image processing steps on a flow cell image. Each FPGA 120 can perform its own portion of the processing step in parallel, thereby reducing the time required to process data, while each AI chip can perform polymerase cloning or cluster prediction after receiving data from its corresponding FPGA. This allows image processing steps to be completed in real time. For example, the first and second FPGAs can be configured to perform intensity registration in parallel on different sub-blocks or blocks of the flow cell. The corresponding AI chip can then perform high-resolution flow cell image prediction on the corresponding sub-block or block after image registration is completed via its respective FPGA. Further discussion of the use of FPGAs is provided below.

[0085] Reconfigurable logic devices and / or integrated circuits can be configured to perform some or all of the operations or actions disclosed herein in real time. Real-time execution of operations or actions allows system 110 to use less memory and / or data storage because data can be processed as it is received. This is an improvement over conventional systems that may require storing data before processing it and therefore require more memory / data storage or access to a computer system located in the cloud 130. Furthermore, real-time execution of operations or actions allows for more efficient sequencing analysis because these operations or actions can be performed in parallel while the sequencing run is still in progress. Moreover, using FPGAs and AI chips to perform processing steps allows the system to use significantly less power, for example, 2x, 5x, 10x, 20x, or more, and thus generate less heat compared to using CPUs and / or GPUs to perform the same processing steps. Further discussion of the use of FPGAs is provided below.

[0086] As discussed above, sequencing system 110 may have a dedicated processor 118, reconfigurable logic devices and / or integrated circuits 120 or computing system 126. The sequencing system may use one, two, or all of these elements to perform one or more operations or actions of the methods disclosed herein. In some embodiments, when these hardware elements are present together, image processing tasks are separated between them. For example, reconfigurable logic device 120 may be used to perform some or all of the following: preprocessing operations, color correction, generation of polymerase clone maps, image registration, prediction of high-resolution flow cell images, training of neural networks, generation of training flow cell images, base interpretation, and any subsequent operations, while computing system 126 may perform other processing functions for sequencing system 110, such as intensity normalization and registration of images used for base interpretation with cell staining images. Those skilled in the art will understand that various combinations of these elements will achieve various system embodiments that balance the efficiency and speed of processing with the cost of the processing elements.

[0087] In some embodiments, one or more reconfigurable logic devices and / or integrated circuits 120 may accelerate base interpretation and / or any primary analysis steps of flow cell images acquired from 2D or 3D samples. In some embodiments, the reconfigurable logic devices and / or integrated circuits may accelerate primary analysis of 2D or 3D volumetric samples by 2x, 4x, 5x, 10x, 15x, 20x, 25x, 30x, 40x, 50x, 100x, 200x, 400x, 500x, 800x, 1000x or more compared to conventional primary analysis methods using only CPUs and / or GPUs. In some embodiments, one or more reconfigurable logic devices and / or integrated circuits 120 herein may accelerate sequencing and sequencing analysis (including at least primary analysis) of flow cell images acquired from 2D or 3D samples. In some embodiments, the reconfigurable logic devices and / or integrated circuits described herein can accelerate sequencing and sequencing analysis (including at least primary analysis) of flow cell images acquired from 2D or 3D samples by 2x, 4x, 5x, 10x, 15x, 20x, 25x, 30x, 40x, 50x, 100x, 200x, 400x, 500x, 800x, 1000x, or more compared to conventional sequencing systems using only CPUs and / or GPUs. In some embodiments, making inferences or predictions about high-resolution images, base interpretation results, or classifications using the neural networks and reconfigurable logic devices and / or integrated circuits disclosed herein can be less than 800 ms, 500 ms, 400 ms, 300 ms, 200 ms, 100 ms, 50 ms, 20 ms, or less per block per cycle. Block sizes in different flow cells can vary. Block sizes can be at least 0.001 mm. 2 0.01 mm 2 0.05 mm 2 0.1 mm 2 0.5 mm 2 1 mm 2 2 mm 2 3 mm 2 Or larger.

[0088] In some embodiments, one or more reconfigurable logic devices and / or integrated circuit 120 can perform primary analysis (base interpretation) of polymerase clones of flow cell images at multiple z-levels. For example, for at least 50 flow cell images (e.g., covering 50 blocks and from two or more color channels), where the FOV is at least 1 mm. 2For one or more flow cycles (e.g., 1 to 15 cycles), the resolution is 1 μm or better in three dimensions, and the processing time using reconfigurable logic devices can be less than 400 hours. Flow cell images can be derived from multiple z-levels to cover some or all of the volumetric 3D samples (e.g., completely covering at least two samples).

[0089] In some embodiments, one or more reconfigurable logic devices and / or integrated circuits 120 can be used to accelerate primary analysis of 3D samples, including training neural networks and using the trained neural networks to make predictions or inferences. For example, the neural network can be used to predict polymerase clone locations and / or predict cell boundaries, thereby identifying polymerase clones within cells. Using reconfigurable logic devices and / or integrated circuits 120 for computations associated with neural networks reduces the required training and / or prediction time compared to using GPUs or other computer processors, thereby accelerating sequence analysis and enabling sequence analysis of flow cycles while they are pending execution or in progress during subsequent flow cycles. In some embodiments, reconfigurable logic devices and / or integrated circuits 120 can accelerate training and / or prediction speeds by 10x, 20x, 50x, 80x, 100x, 200x, 500x, 600x, 800x, 1000x, or higher compared to using CPUs and / or GPUs for training and / or prediction. In some embodiments, reconfigurable logic devices and / or integrated circuits 120 can be used to achieve optimal acceleration of sequence analysis. For example, one or more FPGA chips can be used in combination with an integrated circuit (e.g., an NPU) dedicated to computations corresponding to artificial intelligence (AI) algorithms. The integrated circuit can be a circuit dedicated to AI functionality. The integrated circuit may include an application-specific integrated circuit (ASIC). Computational tasks can be distributed between the FPGA and the integrated circuit to optimize computation time, power consumption, heat dissipation, etc. For example, the AI ​​chip can be used only for computations involving neural networks (e.g., predicting polymerase cloning positions, predicting high-resolution flow cell images, or training neural networks), and the FPGA can be used for the remainder of the primary analysis steps. For at least 50 flow cell images (e.g., approximately 50 blocks covering the flow cell and from two or more color channels), where the field of view (FOV) is at least 1 mm... 2For one or more flow cycles (e.g., 1 to 15 cycles), the resolution of each flow cell image in three dimensions is 1 μm or higher. The initial analysis time using dual FPGA chips or a single FPGA chip combined with an AI chip can be less than 400 hours, 300 hours, 200 hours, 100 hours, 50 hours, or 20 hours. Flow cell images can be derived from multiple z-levels to cover some or all of the volumetric 3D sample (e.g., 10 to 20 z-positions to fully cover at least two samples). The initial analysis time can include the total image processing time from acquiring the raw flow cell images using imager 116 to generating and saving the base interpretation results. The 3D samples in this paper comprise polymerase clones or clusters centered at different z-levels, which are spaced apart from each other by a distance of at least 0.01 μm, 0.05 μm, 0.1 μm, 0.2 μm, 0.5 μm, 1 μm, or greater along the z-direction or axis.

[0090] Cloud 130 can be a network, a remote storage device, or some other remote computing system separated from sequencing system 110. Connection to cloud 130 allows access to data stored outside sequencing system 110 or allows updates to software within sequencing system 110.

[0091] Reconfigurable logic devices and integrated circuits

[0092] Figure 5C Exemplary embodiments of the reconfigurable logic devices and integrated circuits of the sequencing system disclosed herein are shown. In some embodiments, the sequencing system 110 may include one or more reconfigurable logic devices 120_a. Figure 5CIn the illustrated embodiment, the sequencing system includes a single reconfigurable logic device, namely the first reconfigurable logic device 120_a. In some embodiments, the sequencing system includes multiple reconfigurable logic devices (not shown). Each reconfigurable logic device may include a data processing engine 5011 configured to perform data processing in parallel. Each data processing engine may include a combination of digital logic circuits to perform its function, such as intensity extraction, convolution, registration, etc. The sequencing system 110 may further include a reconfigurable routing channel 5013, which can be used as a connection between data processing engines 5011 and can also connect the data processing engines to other structural elements of the sequencing system 110, such as a first processor and memory device. In some embodiments, a neural network may be at least partially deployed on the reconfigurable logic device 120_a, such that the reconfigurable logic device can be used for at least some computational tasks to generate inference using the neural network. The neural network can be pre-trained using various training methods and data, for example, using the training methods and training data disclosed herein. The sequencing system may further include a first processor 120_c, which is used to selectively activate or deactivate different combinations of data processing engine 120_a and reconfigurable routing channels 120_b. The sequencing system 110 is as follows... Figure 1 The FPGA 120 shown may include one or more of a reconfigurable logic device 120_a, an integrated circuit 120_b, and a processor 120_c.

[0093] In some embodiments, FPGA 120 may include only reconfigurable logic device 120_a and processor 120_c, but not integrated circuit 120_b. Different combinations of data processing engine 5011 and reconfigurable routing channel 5013 may be configured to perform operations in sequencing analysis to facilitate the generation of sequencing analysis results. Sequencing analysis may include operations or steps of primary analysis. Such operations may include one or more of the following: (a) acquiring sensor data from one or more sensors of the sequencing system (located in imager 116); (b) processing the sensor data to generate a first plurality of flow cell images; (c) using a neural network to predict a second plurality of flow cell images based on the sensor data or the first plurality of flow cell images; (d) identifying polymerase clones from the second plurality of flow cell images; and (e) performing a corresponding base readout on each of the identified polymerase clones based on the second plurality of flow cell images.

[0094] In some embodiments, sensor data includes raw data acquired from the imager's sensors without any additional image processing. In some embodiments, sensor data includes raw flow cell images that have not yet been processed by the computing system 126, dedicated processor 118, and / or reconfigurable logic devices and integrated circuits 120 of the sequencing system 110.

[0095] In some embodiments, the sequencing system includes: a first reconfigurable logic device 120_a, comprising a first plurality of data processing engines 5011 configured to perform data processing in parallel; a first reconfigurable routing channel 5013 connected to at least some of the first plurality of data processing engines 5011; a neural network at least partially deployed on the first reconfigurable logic device 5011; and a first processor 120_c, which selectively activates or deactivates different combinations of the first plurality of data processing engines 5011 and the first reconfigurable routing channel 5013 to perform operations in sequencing analysis, thereby facilitating the generation of sequencing analysis results. Sequencing analysis may include operations or steps of primary analysis. Such operations may include one or more of the following: (a) obtaining sensor data directly from one or more sensors of a sequencing system; (b) processing the sensor data to generate a first plurality of flow cell images; (c) performing a first convolution on the first plurality of flow cell images in one or more dimensions, thereby generating a first convolution result; (d) repeating a downsampling operation including the following once or more: (1) performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result; and (2) performing downsampling on the second convolution result according to a downsampling factor, thereby generating a first downsampled result, wherein in each repetition, the second convolution includes a corresponding number of filters, thereby generating a third convolution result after (d); (e) performing a second convolution on the third convolution result in one or more dimensions, thereby generating a fourth convolution result; (f) repeatedly performing an upsampling operation including the following: (3) performing upsampling on the fourth convolution result according to an upsampling factor, thereby generating a first upsampled result; and (4) performing downsampling on one or more dimensions. The first upsampling result is subjected to a second convolution, thereby generating a fifth convolution result, wherein in each repetition, the second convolution includes a corresponding number of filters, thereby generating a sixth convolution result after (f); (g) the sixth convolution result is subjected to a first convolution in one or more dimensions, thereby generating a seventh convolution result; (h) based on the seventh convolution result, a second plurality of flow cell images are predicted, wherein each of the second plurality of flow cell images corresponds to a corresponding flow cell image in the first plurality of flow cell images with a second resolution, the second resolution being at least 2, 4, 6, 8, 10, 12, 16, or 32 times larger than the first resolution in one or more spatial dimensions; (i) polymerase clones in the second plurality of flow cell images are identified; (j) based on the second plurality of flow cell images, a corresponding base readout is performed on each of the identified polymerase clones; and (k) optionally, the second plurality of flow cell images, the corresponding base readout, or both are sent to a first reconfigurable logic device, a first processor, or one or more hardware processors of the sequencing system.

[0096] In some embodiments, sensor data can be acquired from one or more sensors (located in imager 116) of the sequencing system via a direct connection. In some embodiments, the direct connection between the first reconfigurable logic devices (120 and 120_a) and the sensors lacks other hardware components that could process or store the sensor data, thus introducing unnecessary complexity, latency, and potential errors in sensor data communication. Such hardware components include the first processor 120_c, memory device 5030, or any processor, such as the computing system 126 of the sequencing system, for example, a CPU. Compared to conventional sequencing systems where sensor data is first transferred to other hardware components, such as the CPU, before being transferred to the GPU for processing, the direct sensor data communication described herein advantageously improves the data transfer efficiency from the sensor to the FPGA 120, frees up other hardware (e.g., CPU, storage device) for other data processing functions, reduces the power consumption of indirect data communication, and reduces the time consumed by data communication, thereby shortening the sequencing analysis time.

[0097] In some embodiments, the connection between the first reconfigurable logic devices (120 and 120_a) and the sensor may include other hardware components that can process or store sensor data. Such hardware components may include a first processor 120_c, a memory device 5030, or any processor, such as the CPU 126 of a sequencing system. For example, sensor data may be stored in the memory device 5030 and then accessed by the first reconfigurable logic devices via the storage controller 5013.

[0098] Reconfigurable logic devices may include digital logic circuits, and in a sense, they are also integrated circuits. However, integrated circuits (e.g., AI chips, NPUs, etc.) as described herein can differ in various ways from reconfigurable logic devices; for example, integrated circuits may not offer the same reconfiguration flexibility as reconfigurable logic devices. For instance, integrated circuits (e.g., AI chips, NPUs, etc.) as described herein may be non-reconfigurable.

[0099] In some embodiments, the sequencing system 110 includes at least one reconfigurable logic device, but lacks any integrated circuits, such as an AI chip, an ASIC chip, or an NPU. The reconfigurable logic device can perform one or more operations in the sequencing analysis and can send its output back to the CPU as the final result of a primary analysis (e.g., base interpretation results). Alternatively, the reconfigurable logic device can send its output back to the CPU, allowing subsequent operations to be performed by the CPU based on its output to generate the final result of the sequencing analysis.

[0100] In some embodiments, the sequencing system 110 includes at least one reconfigurable logic device and at least one integrated circuit, such as Figure 5C As shown in the diagram, the integrated circuit can perform one or more operations in sequencing analysis and can send its output back to a reconfigurable logic device, allowing subsequent operations to be performed at the reconfigurable logic device based on its output.

[0101] In some embodiments, the output of the reconfigurable logic device or integrated circuit includes the base identification results of nucleotide bases in a sample immobilized on a support. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base identification positions in two dimensions. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base identification positions in three dimensions.

[0102] In some embodiments, data communication between any two of the reconfigurable logic device, integrated circuit, first processor, and second processor may be direct, such that direct communication does not involve any other hardware components capable of processing or storing data. Such other hardware components may include memory devices and / or other processors in the sequencing system. Such direct communication may include DMA connections. In some embodiments, data communication between any two of the reconfigurable logic device, integrated circuit, first processor, and second processor may be direct, such that data is not used or stored by other logic circuitry before reaching its communication destination, but data may be stored in a memory device before reaching its communication destination.

[0103] In some embodiments, the sequencing system 110 may include: a first reconfigurable logic device 120_a, such as an FPGA, including a first plurality of data processing engines 5011 configured to perform data processing in parallel; an integrated circuit 120_b, such as an AI chip; a neural network at least partially deployed on the integrated circuit; and a first processor that selectively activates or deactivates different combinations of the first plurality of data processing engines, used individually or in combination with a first routing channel, to perform operations in sequencing analysis to facilitate the generation of sequencing analysis results. Sequencing analysis may include operations or steps of primary analysis. Sequencing analysis may include operations or steps of secondary analysis. Such operations may include one or more of the following: acquiring sensor data from one or more image sensors of the sequencing system; processing the sensor data to generate a first plurality of flow cell images; and transmitting the sensor data, the first plurality of flow cell images, or both, to the integrated circuit. The sequencing system may include a second processor or a first processor to control the integrated circuit to perform one or more operations in sequencing analysis to facilitate the generation of sequencing analysis results. Sequencing analysis may include operations or steps of primary analysis and / or secondary analysis. Such operations may include one or more of the following: receiving sensor data, a first plurality of flow cell images, or both, from a first reconfigurable logic device; using a neural network to predict a second plurality of flow cell images based on the sensor data, the first plurality of flow cell images, or both; identifying polymerase clones from the second plurality of flow cell images; performing a corresponding base readout on each of the identified polymerase clones based on the second plurality of flow cell images; and sending the second plurality of flow cell images, the identified polymerase clones, and the corresponding base readouts of the polymerase clones in the second plurality of flow cell images to one or more of the following: the first reconfigurable logic device 120_a, the first processor 120_c or the second processor, and / or one or more processors of the sequencing system 126. In some embodiments, sending a second plurality of flow cell images and identified polymerase clones; the operation of base reading of the corresponding polymerase clones in the second plurality of flow cell images includes sending to a memory device herein, such as DDR memory, such that one or more of a first reconfigurable logic device 120_a, a first processor 120_c or a second processor, and / or one or more processors of computing system 126 can access data from the memory. Access to data from the memory (including reading, writing, editing, etc.) may be assisted by a memory controller disclosed herein.

[0104] In some embodiments, the sequencing system 110 includes at least one reconfigurable logic device and at least one integrated circuit, such as Figure 5CAs shown in the diagram. The integrated circuit can perform one or more operations in sequencing analysis and can generate the final result of the primary analysis from its output and send the output to one or more devices, including: a reconfigurable logic device, a first or second processor, a hardware processor of the sequencing system, etc., so that the final result can be saved or presented to a user. In some embodiments, the output of the reconfigurable logic device or integrated circuit includes the base reading results of nucleotide bases in a sample immobilized on a support. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base reading positions in two dimensions. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base reading positions in three dimensions.

[0105] In some embodiments, the integrated circuit can perform one or more operations in the sequencing analysis and generate intermediate results of the primary analysis, such as the location of the polymerase clone, and can send its output back to one or more of the following: a reconfigurable logic device, a first or second processor, a hardware processor of the sequencing system, etc., so that the final result can be determined based on its output.

[0106] In some embodiments, the integrated circuit may send its output (intermediate or final result) to be stored in a memory device, such that one or more devices (including: a reconfigurable logic device, a first or second processor, and a hardware processor of the sequencing system) can access the stored output when needed. Access to the output stored in the memory device may be made by a storage controller (e.g., 5013) of the sequencing system.

[0107] In some embodiments, the output of the reconfigurable logic device or integrated circuit includes the base identification results of nucleotide bases in a sample immobilized on a support. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base identification positions in two dimensions. In some embodiments, the output data of the reconfigurable logic device or integrated circuit includes the identification of base identification positions in three dimensions.

[0108] In some embodiments, the sequencing system includes: a first reconfigurable logic device including a first plurality of data processing engines configured to perform data processing in parallel with each other; an integrated circuit; a neural network at least partially deployed on the integrated circuit; and a first processor selectively activating or deactivating different combinations of the first plurality of data processing engines. The different combinations of the first plurality of data processing engines may be configured to perform operations including: acquiring sensor data from one or more image sensors of the sequencing system to generate a first plurality of flow cell images; and transmitting the sensor data, the first plurality of flow cell images, or both, to the integrated circuit. The integrated circuit may perform operations including: (1) receiving sensor data, the first plurality of flow cell images, or both, from the first reconfigurable logic device; and (2) predicting a second plurality of flow cell images using the neural network based on the sensor data, the first plurality of flow cell images, or both; and (3) transmitting the second plurality of flow cell images to the first reconfigurable logic device or one or more hardware processors of the sequencing system.

[0109] In some embodiments, the sequencing system includes: a first reconfigurable logic device including a first plurality of data processing engines arranged in a first pipeline and configured to perform data processing in parallel with each other; an integrated circuit; a neural network at least partially deployed on the integrated circuit; and a first processor of the first reconfigurable logic device for selectively activating or deactivating different combinations of the first plurality of data processing engines to perform operations including: (a) acquiring sensor data from one or more sensors of the sequencing system; (b) processing the sensor data to generate a first plurality of flow cell images; and (c) transmitting the sensor data, the first plurality of flow cell images, or both, to the integrated circuit; wherein the integrated circuit performs operations including: (d) receiving sensor data, the first plurality of flow cell images, or both, from the first reconfigurable logic device; (e) performing a first convolution on the first plurality of flow cell images in one or more dimensions, thereby generating a first convolution result; and (f) repeating a downsampling operation including: (1) performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result; and (2) performing a downsampling operation including: (a ...b) performing a downsampling operation including: (c) performing a second convolution on the first convolution result in one or more dimensions; and (d) performing a downsampling operation including: (a) performing a second convolution on the first convolution result in one or more dimensions; and (c) performing a downsampling operation including: (a) performing a second convolution on the first convolution result in one or more dimensions; and (d) performing a downsampling operation on the first reconfigurable logic device. (g) Perform downsampling on the second convolution result according to the downsampling factor, thereby generating a first downsampling result, wherein in each repetition, the second convolution includes the corresponding number of filters, thereby generating a third convolution result; (h) Perform a second convolution on the third convolution result in one or more dimensions, thereby generating a fourth convolution result; (g) Repeatedly perform upsampling operations including: (3) Performing upsampling on the fourth convolution result according to the upsampling factor, thereby generating a first upsampling result; and (h) Performing a second convolution on the first upsampling result in one or more dimensions, thereby generating a fifth convolution result. The result, wherein in each repetition, the second convolution includes a corresponding number of filters, thereby generating a sixth convolution result; (i) performing a first convolution on the sixth convolution result in one or more dimensions, thereby generating a seventh convolution result; and (j) predicting a second plurality of flow pool images based on the seventh convolution result, wherein each of the second plurality of flow pool images corresponds to a corresponding flow pool image in the first plurality of flow pool images having a second resolution, the second resolution being at least 2, 4, 6, 8, 10, 12, or 16 times larger than the first resolution in one or more spatial dimensions.

[0110] In some embodiments, the first reconfigurable routing channel includes one or more electronic nodes, and these electronic nodes are programmable. Electronic nodes here may include connection points in a circuit. An electronic node may include a point where two or more circuit elements are connected together. In some embodiments, the first reconfigurable routing channel includes one or more interconnects. Interconnects may include physical wiring connecting transistors and other components on an integrated circuit. In some embodiments, the reconfigurable routing channel includes one or more memory controllers, for example, Figure 5C5013 in the example. In some embodiments, the first reconfigurable routing channel includes one or more on-chip networks (NoC), for example, Figure 5C 5013 in the example. In some embodiments, the first reconfigurable routing channel may include one or more of the following: a network on-chip (NoC) and a memory controller.

[0111] The first reconfigurable routing channel can be configured to passively transfer data between components of the sequencing system. For example, the reconfigurable routing channel can be configured in a data processing engine (e.g., Figure 5C 5011 in the middle) and memory devices (e.g., Figure 5C The first reconfigurable routing channel can be configured to allow data communication between the first reconfigurable logic device (e.g., 120_a) and one or more memory devices (e.g., 5030). The first reconfigurable routing channel can also be configured to allow data communication between the first reconfigurable logic device (e.g., 120_a) and an integrated circuit (e.g., 120_b).

[0112] The reconfigurable logic device described herein may include one or more data processing engines (e.g., 5011). Each data processing engine may include multiple digital logic circuits.

[0113] The first reconfigurable logic device may be configured to communicate data with one or more external memory devices. The first reconfigurable logic device may be configured to transfer data to one or more external memory devices via a first reconfigurable routing channel. The first reconfigurable logic device may include digital circuitry that is integrated and forms an FPGA device. For example, Figure 5C The FPGA device in the FPGA includes a first reconfigurable logic device, a DMA connection, and a first reconfigurable routing channel (e.g., NoC and memory controller).

[0114] The sequencing system may further include one or more memory devices electrically connected to communicate data with one or more components of the sequencing system, the one or more components including one or more of the following: a first reconfigurable logic device; an integrated circuit; a first reconfigurable routing channel; one or more memory controllers; a first processor; a second processor; and one or more processors of the sequencing system.

[0115] In some embodiments, the sequencing system further includes one or more direct data access (DMA) connections, for example, Figure 5C In the 5012, these connections are with multiple data processing engines and the first reconfigurable routing channel (e.g., Figure 5CData communication is performed between components of the sequencing system (e.g., 5013). DMA connections can be configured to actively communicate data between components of the sequencing system. For example, a DMA connection can be configured to transmit data from a component it is connected to (e.g., 5013). Figure 5C 5011 and Figure 5C (5013) Acquires or sends data to. The DMA connection described herein can be configured to actively request data from or actively send data directly to: a first reconfigurable logic device; a first reconfigurable routing channel; an integrated circuit; or a combination thereof. One or more Direct Data Access (DMA) connections can communicate with the first reconfigurable routing channel and the integrated circuit described herein. The DMA connection can be configured to allow data communication based on a predetermined protocol (e.g., the PCIe protocol).

[0116] In some embodiments, the first reconfigurable routing channel is configured to allow data communication between the first reconfigurable logic device and one or more memory devices. In some embodiments, the one or more DMA connections and the first reconfigurable routing channel are configured to allow data communication between the first reconfigurable logic device and an integrated circuit.

[0117] In some embodiments, the sequencing system further includes an integrated circuit, which is connected to a first reconfigurable logic device (e.g., Figure 5C This differs from 120_b). The integrated circuit described herein may be non-reconfigurable. The integrated circuit may include an application-specific integrated circuit (ASIC) chip. In some embodiments, the integrated circuit includes a neural processing unit (NPU) or an artificial intelligence (AI) chip. The integrated circuit may include a second plurality of data processing engines, each data processing engine including a plurality of digital logic circuits. The integrated circuit may further include: a second plurality of data processing engines and a second routing channel, each routing channel connecting at least some of the second plurality of data processing engines.

[0118] In some embodiments, the sequencing system further includes a first processor. The first processor is configurable to selectively activate or deactivate different combinations of a first plurality of data processing engines and a first reconfigurable routing channel to perform the operations disclosed herein. In some embodiments, the sequencing system further includes a second processor. The second processor is configurable to control the digital circuitry of the integrated circuits herein.

[0119] In some embodiments, a first or second processor (e.g., an integrated circuit) is configured to selectively activate or deactivate different combinations of a second plurality of data processing engines and a second reconfigurable routing channel to perform operations. The first or second processor may be configured to selectively activate or deactivate different combinations of a second plurality of data processing engines and a second reconfigurable routing channel to perform the operations described herein.

[0120] The sequencing system may further include a housing that encloses a first reconfigurable logic device, a first reconfigurable routing channel, the one or more DMA connections, an integrated circuit, and a first processor. In some embodiments, the sequencing system further includes a housing that encloses at least the first reconfigurable logic device, and the integrated circuit is located outside the housing.

[0121] In some embodiments, the sequencing system further includes a power supply configured to provide different power levels to the first reconfigurable logic device and the integrated circuit. While sequencing runs and / or sequencing analyses are in progress, a first power level supplied to the first reconfigurable logic device may be higher than a second power level supplied to the integrated circuit. The maximum output power of the sequencing system's power supply is 2, 3, 5, 8, 10, or 20 times lower than the maximum output power of a sequencer (e.g., a conventional sequencer) that does not include the first reconfigurable logic device (e.g., an FPGA), the integrated circuit (e.g., an AI chip), or both. The time consumed by performing a sequencing run and its corresponding sequencing analysis (e.g., a primary analysis) using the sequencing system is 2, 3, 5, 8, 10, or 20 times shorter than the time consumed by performing the same sequencing run using a sequencer that does not include the first reconfigurable logic device, the integrated circuit, or both (e.g., a conventional sequencer that does not include an FPGA and / or AI chip). The time consumed by using a sequencing system to perform a sequencing run and the sequencing analysis (e.g., a primary analysis) of that sequencing run is 2, 3, 5, 8, 10, or 20 times shorter than the time consumed by using a sequencer that does not include a first reconfigurable logic device, an integrated circuit, or both (e.g., a conventional sequencer that does not include an FPGA and / or an AI chip) to perform the same sequencing run and analysis. In some embodiments, the maximum output power supplied to the sequencing system by the power supply during the execution of the sequencing run and its corresponding sequencing analysis is less than 900 watts, 800 watts, 700 watts, 650 watts, 600 watts, 550 watts, or 500 watts. The power supply may be configured to supply a first power level to the first reconfigurable logic device, the first power level being less than 500 watts, 400 watts, 350 watts, or 300 watts. The power supply may be configured to supply a second power level to the integrated circuit, the second power level being less than 450 watts, 400 watts, 350 watts, or 300 watts.

[0122] In some embodiments, one or more components of the first reconfigurable logic device and / or integrated circuit may include a computing performance of at least 2, 4, 8, 10, 16, 20, 30, 40, 50, 60, 70, 80, or 100 gigaoperations per second (GOPs) or higher. In some embodiments, one or more processing engines of the first reconfigurable logic device and / or integrated circuit may include a computing performance of at least 12, 4, 8, 10, 16, 20, 30, 40, 50, 60, 70, 80, or 100 gigaoperations per second (GOPs) or higher. In some embodiments, the first reconfigurable logic device and / or integrated circuit includes a computing performance of at least 10, 20, 40, 50, 60, 80, or 100 trillion operations per second (TOPs).

[0123] In some embodiments, one or more components are located on a first printed circuit board (PCB). These components may include: a first reconfigurable logic device; a first reconfigurable routing channel; a first processor; and one or more DMA connections. In some embodiments, the integrated circuit is located on a second printed circuit board (PCB) different from the first printed circuit board, for example, as shown below. Figure 5C As shown in the diagram. The integrated circuit and the second PCB may be located within the same sequencing system housing as the first PCB, or outside the sequencing system housing. Arranging them on separate PCBs allows for convenient, efficient, and easily customized interconnection of the first reconfigurable logic device (e.g., an FPGA device) with various integrated circuits on the chip. In some embodiments, the first PCB may be a motherboard, and the second PCB may be a daughterboard or an edge unit.

[0124] In some embodiments, the sequencing system lacks any graphics processing unit (GPU) or tensor processing unit (TPU). Instead, the sequencing system utilizes an FPGA, AI chip, NPU, or other ASIC chip to perform the operations disclosed herein. Compared to sequencers using GPUs or TPUs, the sequencing system disclosed herein advantageously requires less power, generates less heat, and reduces the hardware complexity and cost for performing NGS sequencing runs and corresponding sequencing analyses.

[0125] In some embodiments, the sequencing system includes logic devices, which are not limited to reconfigurable logic devices (e.g., FPGAs) and / or other integrated circuits (e.g., AI chips, NPUs). In some embodiments, the sequencing system includes various types of processing units or processors configured for reconfigurable parallel processing. In some embodiments, the sequencing system includes various types of logic devices or integrated circuits, such as ASIC chips. In some embodiments, the sequencing system includes GPUs, TPUs, or other various types of processing units configured to perform one or more of the operations disclosed herein.

[0126] In some embodiments, the sequencing system includes a GPU, TPU, or other processing unit of various types configured to perform one or more operations that can be performed by reconfigurable logic devices (e.g., FPGAs) and / or other integrated circuits (e.g., AI chips, NPUs).

[0127] A first processor may be located on a first PCB board along with a reconfigurable logic device for convenient and efficient control of the reconfigurable logic device. In some embodiments, the first processor is a separate processor from one or more processors of the sequencing system, and the first processor is configured to control an optical system, fluids of the sequencing system, etc. In some embodiments, the first processor may be configured to control only components on the first PCB board (e.g., an FPGA device), used alone or in combination with components on a second PCB board (e.g., an AI chip). In some embodiments, the sequencing system may include a second processor configured to control the AI ​​chip independently. The first or second processor of the sequencing system (e.g., 120_c) may include a CPU. The one or more hardware processors of the sequencing system include a CPU. In some embodiments, the first or second processor (e.g., 120_c) lacks any GPU or TPU. In some embodiments, the first or second processor (e.g., 120_c) includes only a CPU.

[0128] In some embodiments, the sequencing system may further include a heat sink configured to maintain the system temperature in the range of 0°C to 120°C or below 120°C.

[0129] In some embodiments, the operation of processing sensor data to generate a first plurality of flow cell images includes one or more of the following: registering the first plurality of flow cell images to a reference coordinate system; adjusting the image intensity of the first plurality of flow cell images; performing color correction on the first plurality of flow cell images; performing hysteresis and advance correction on the first plurality of flow cell images; and subtracting the background intensity from the first plurality of flow cell images.

[0130] In some embodiments, each of the one or more operations performed by the first reconfigurable logic device or integrated circuit is a real-time operation. In some embodiments, each of the one or more operations performed by the first reconfigurable logic device or integrated circuit is performed within a time window for executing a sequencing reaction and / or imaging a single sequencing cycle. In some embodiments, each of the one or more operations performed by the first reconfigurable logic device or integrated circuit is performed within a time window for executing a sequencing reaction and / or imaging a single z-level of a single sequencing cycle.

[0131] Figure 5DAn exemplary embodiment of performing sequencing analysis in parallel with the sequencing run is illustrated. In this particular embodiment, the sequencing run comprises multiple sequencing cycles, and only a portion of a single cycle is shown herein. For each cycle, flow cell images are acquired at multiple z-levels from different color channels of the in-situ sample. Within each cycle, the sequencing reaction is repeated for each z-level within a time window 5601. Operation of the integrated circuit is performed within a processing window 5602 within a time window 5609 of the single sequencing cycle, and also within a time window 5601 for the sequencing reaction and imaging at the single z-level 5601. Operation of the first reconfigurable logic device (e.g., onboard primary analysis operation) is also performed within a processing window 5603 within a time window 5609 of each sequencing cycle. The durations of processing windows 5602 and 5603 may be the same or different, depending on various factors such as sequencing data, primary analysis algorithms, etc. In some embodiments, operations are performed not only within the processing window but also completed within the processing window for data of the current cycle, for example, for data of the previous z-level of the current cycle for acquired sensor data. In some embodiments, the operation is performed within a processing window on data from the previous cycle, for example, the cycle immediately preceding the current cycle.

[0132] In embodiments where the sample is a 3D volume sample, these operations are performed within a predetermined time window (e.g., 5602, 5603) within a single z-level of each cycle. The predetermined time window is for a single z-level in a single sequencing cycle. In some embodiments, the predetermined time window is less than 1000 ms, 900 ms, 800 ms, 700 ms, 600 ms, 500 ms, 400 ms, 300 ms, 250 ms, 200 ms, or 100 ms. In some embodiments, each of the one or more operations is performed in parallel within the predetermined time window while the sequencing run is in progress. In some embodiments, each of the one or more operations is performed in parallel within a time window after sequencing, imaging, or both are completed in a subsequent sequencing cycle.

[0133] The first plurality of flow cell images described herein may be obtained from a single z-level of a 2D or 3D sample. In some embodiments, the first plurality of flow cell images described herein may be obtained from multiple z-levels covering at least a portion of the in situ sample (e.g., cells or tissue). The first plurality of flow cell images may be obtained from one or more color channels of each z-level of the multiple z-levels covering at least a portion of the in situ sample. In some embodiments, the first plurality of flow cell images are derived from a single color channel. In some embodiments, the first plurality of flow cell images are derived from multiple color channels. In some embodiments, the first plurality of flow cell images are derived from a single sequencing cycle. In some embodiments, the first plurality of flow cell images are derived from multiple sequencing cycles. The first plurality of flow cell images may have a first spatial resolution in the x, y, and / or z directions. A second plurality of flow cell images may be generated based on the first plurality of flow cell images. The second plurality of flow cell images may have a second spatial resolution in the x, y, and / or z directions. The first spatial resolution may be lower than the second spatial resolution, and the higher resolution described herein indicates a smaller pixel size, resulting in finer spatial detail of the polymerase clones in the flow cell images. The first spatial resolution may be 2, 4, 6, 8, 10, 16, 24, 32, or 48 times lower than the second spatial resolution in the x, y, and / or z directions. The first spatial resolution may be at least 2, 4, 6, 8, 10, 16, 24, 32, or 48 times lower than the second spatial resolution in the x, y, and / or z directions. In some embodiments, the first and second resolutions are 3D resolutions. In some embodiments, the first resolution ranges from 0.1 μm to 5 μm. In some embodiments, the second resolution ranges from 0.01 μm to 2 μm. In some embodiments, the second resolution is at least 4, 6, or 8 times larger than the first resolution in all three dimensions.

[0134] In some embodiments, the sequencing system further includes one or more image sensors configured to receive optical signals generated by a sequencing reaction from a sample fixed on a support. The support may include a glass or plastic substrate. The support may be included in a flow cell device. The one or more image sensors may be configured to generate sensor data based on the optical signals. In some embodiments, the sequencing system further includes: one or more hardware processors; and one or more data storage devices storing instructions executable by the one or more hardware processors to cause the one or more hardware processors to perform the operations disclosed herein. The one or more data storage devices may include one or more memory devices. The one or more memory devices may be accessible by the one or more processors, a first processor, a second processor, a first reconfigurable logic device, and an integrated circuit.

[0135] In some embodiments, the one or more processors are separate from the first processor or the second processor. Operations performed by the one or more processors may include one or more of the following: 1) recording sensor data generated in the sequencing system during one or more flow cycles; 2) optionally processing the recorded sensor data; 3) sending the recorded sensor data or optionally processed data to a first reconfigurable logic device or integrated circuit; 4) receiving output from the first reconfigurable logic device or integrated circuit; and 5) generating sequencing analysis results based on the received output. Operations performed by the one or more processors may include one or more of the following: 1) receiving output from the first reconfigurable logic device or integrated circuit; and 2) generating sequencing analysis results based on the received output.

[0136] In some embodiments, sequencing analysis results include preliminary analysis results. In some embodiments, sequencing analysis results include data files in a predetermined data format. In some embodiments, sequencing analysis results include base interpretation results of nucleotide bases in a sample immobilized on a support. In some embodiments, sequencing analysis results include quality measurement results of the base interpretation results of nucleotide bases in a sample immobilized on a support. In some embodiments, sequencing analysis results include quality scores corresponding to the base interpretation results of nucleotide bases in a sample immobilized on a support.

[0137] In some embodiments, the sequencing system further includes: a sample fixed on a support; and an optical system including: an illumination system; an objective lens; and one or more image sensors. The optical system is configured to emit light toward the sample and collect light signals emitted from the sample, thereby generating a first plurality of flow cell images. The support may be included in a flow cell device.

[0138] In some embodiments, the operations performed using a neural network via a first reconfigurable routing channel or integrated circuit include one or more of the following: generating quality measurement results of base interpretation; and generating a data output file based on the base interpretation.

[0139] In some embodiments, the neural network of this invention includes a convolutional neural network (CNN). In some embodiments, the neural network includes a U-Net. In some embodiments, the neural network has been pre-trained. In some embodiments, the neural network has been trained using a first reconfigurable logic device or integrated circuit. In some embodiments, the neural network is a 3D neural network.

[0140] In some embodiments, the first convolution includes a 3D convolution with a kernel. In some embodiments, the kernel has at least four dimensions. In some embodiments, the kernel is m x m x m x n, where m is an integer ranging from 3 to 30, and n is an integer. In some embodiments, n is an integer ranging from 1 to 16384. In some embodiments, the second convolution in operation (1) includes corresponding numbers of n, 2, ..., n in the first repetition, the second repetition, the third repetition, and the fourth repetition, respectively. n、4 n and 8 n filters. In some embodiments, the second convolution in (4) includes the corresponding number of 2 in the last repetition, the penultimate repetition, the penultimate repetition, and the penultimate repetition, respectively. n、2 n、4 n and 8 n filters. In some embodiments, n ranges from 4 to 1024.

[0141] In some embodiments, the neural network has been trained using a first reconfigurable logic device or integrated circuit. In some embodiments, the neural network is a 2D neural network. In some embodiments, the first convolution includes a 2D convolution with a kernel. In some embodiments, the kernel has at least three dimensions. In some embodiments, the kernel is m x m x n, where m is an integer ranging from 3 to 30, and n is an integer. In some embodiments, n is an integer ranging from 1 to 16384.

[0142] In some embodiments, the second convolution in operation (1) includes corresponding numbers n, 2, ..., 2 in the first repetition, the second repetition, the third repetition, and the fourth repetition, respectively. n、4 n and 8 n filters. In some embodiments, the second convolution in (4) includes the corresponding number of 2 in the last repetition, the penultimate repetition, the penultimate repetition, and the penultimate repetition, respectively. n、2 n、4 n and 8 n filters. In some embodiments, n ranges from 4 to 1024.

[0143] In some embodiments, the second convolution in operation (1) includes corresponding numbers n, 2, and 2 in the first repetition, second repetition, and third repetition, respectively. n、4 n filters. In some embodiments, the second convolution in (4) includes the corresponding number of 2 in the last repetition, the penultimate repetition, and the penultimate repetition, respectively. n、2 n、4 n filters. In some embodiments, n ranges from 4 to 1024.

[0144] In some embodiments, the neural network is pre-trained using multiple z-level 2D flow cell images that encompass the 3D volume of the volumetric sample. Compared to neural networks trained using training data from 3D volumes (e.g., 3D CNNs), neural networks pre-trained using 2D flow cell images (e.g., 2D CNNs) are less complex and require less computation to make predictions or inferences, thus providing greater efficiency in predicting polymerase clone locations, saving both time and computation. In some embodiments, a neural network pre-trained using 2D flow cell images can predict the polymerase clone location for each block per cycle within a time window that is 10, 50, 80, 100, 200, 400, 600, 800, 1000, 1500, 2000, or less smaller than a neural network trained using 3D volumes from flow cell images to make the same prediction.

[0145] In some embodiments, a neural network pre-trained using 2D flow cell images can use reconfigurable logic devices and / or other integrated circuits (e.g., FPGAs and / or AI chips) to predict the polymerase clone position of each block per cycle within a time window that is 5, 10, 20, 40, 50, 80, 100, 200, 400, 600, 800, 1000, or less smaller than the same neural network using a CPU or other processor.

[0146] In some embodiments, operation (e) of performing corresponding base readings on each of the identified polymerase clones based on a second plurality of flow cell images includes: performing corresponding base readings on each of the identified polymerase clones based on the second plurality of flow cell images and on a fourth plurality of flow cell images, wherein the fourth plurality of images are predicted using a second neural network based on a third plurality of flow cell images. In some embodiments, the third plurality of flow cell images are acquired from one or more color channels different from a single channel, and wherein the third plurality of flow cell images include a first resolution. In some embodiments, the fourth plurality of flow cell images include a second resolution.

[0147] In some embodiments, the first plurality of flow cell images are derived from one or more color channels. In some embodiments, the first plurality of flow cell images exhibit unbalanced nucleotide diversity. In some embodiments, the first plurality of flow cell images include: unbalanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more flow cycles. In some embodiments, the first plurality of flow cell images include: balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more cycles. In some embodiments, two or more different multiply molecules have different insert sequences. In some embodiments, the different insert sequences correspond to different target RNA molecules or target cDNA molecules. In some embodiments, each position of the identified polymerase clone corresponds to a position of the multiply molecule. In some embodiments, the first plurality of flow cell images include optical signals emitted by nucleotide reagents bound to balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support. In some embodiments, the first plurality of flow cell images include: optical signals emitted by nucleotide reagents that bind to the unbalanced diversity of nucleotide bases A, G, C, and T / U in the polymolecule immobilized on a support, in one or more subsequent cycles. In some embodiments, the unbalanced diversity of nucleotide bases A, G, C, and T / U in the polymolecule includes: (1) the percentage of the number of one or more types of nucleotide bases relative to (2) the total number of bases is less than 20%, 15%, 10%, or 5% in the one or more cycles. In some embodiments, the balanced diversity of nucleotide bases A, G, C, and T / U in the plurality of polymolecules includes: (1) the percentage of the number of each type of nucleotide base relative to (2) the total number of bases is greater than 10%, 15%, or 20% in one or more cycles.

[0148] In some embodiments, the cell sample contains a spatial density ranging from 102 2 Up to 10 15 Each mm 2 Overloaded multiply molecules. In some embodiments, the cell sample contains a spatial density ranging from 102 3 Up to 10 10 Each mm 2 Overloaded multiply molecules.

[0149] In some embodiments, the downsampling factor is 2, 4, or 8. In some embodiments, the upsampling factor is 2, 4, or 8. In some embodiments, the downsampling factor is 2, 4, 8, 16, 32, 64, or greater. In some embodiments, the upsampling factor is 2, 4, 8, 16, 32, 64, or greater.

[0150] In some embodiments, one or more of operations (a) through (k) are performed concurrently with a sequencing run. In some embodiments, the first plurality of flow cell images are acquired in sequencing cycles ranging from 1 to 500. In some embodiments, the one or more cycles include a current cycle N. In some embodiments, N ranges from 1 to 500. In some embodiments, the one or more cycles include a single cycle ranging from 1 to 500. In some embodiments, the one or more cycles include a plurality of cycles ranging from 1 to 500. In some embodiments, one or more of operations (e.g., operations (a) through (j)) are performed concurrently with sequencing reactions in cycles following the current cycle N, even if such reactions have not yet been performed or are being performed.

[0151] In some embodiments, the training dataset for flow pool images comprises a z-stack of flow pool images acquired at different z-positions, and each z-stack is used as a 3D volume for training the neural network. In some embodiments, the training dataset for flow pool images comprises 2D flow pool images acquired at different z-positions, and a single 2D flow pool image at multiple z-levels is used as a 2D image for training the neural network.

[0152] In some embodiments, the training dataset for the flow cell images includes simulated flow cell images of in-situ samples at different z-positions. In some embodiments, the training dataset for the flow cell images includes actual flow cell images obtained from in-situ samples at different z-positions.

[0153] In some embodiments, performing a first convolution on the first plurality of flow pool images in one or more dimensions includes: performing a first convolution on the first plurality of flow pool images in 3D, thereby generating a first convolution result. In some embodiments, performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result, includes: performing a second convolution on the first convolution result in 3D, thereby generating a second convolution result.

[0154] In some embodiments, performing a first convolution on the first plurality of flow pool images in one or more dimensions includes: performing a first convolution on the first plurality of flow pool images in 2D, thereby generating a first convolution result. In some embodiments, performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result, includes: performing a second convolution on the first convolution result in 2D, thereby generating a second convolution result.

[0155] In some embodiments, repeatedly performing the upsampling operation includes: (3) performing upsampling on the fourth convolution result according to the upsampling factor, thereby generating a first upsampling result; (4) concatenating the first upsampling result in the current upsampling repetition with the first downsampling result in the previous downsampling repetition, wherein the first upsampling result and the first downsampling result in the previous downsampling repetition have the same size; and (5) performing a second convolution on the first upsampling result in one or more dimensions, thereby generating a fifth convolution result.

[0156] In some embodiments, different combinations of the first plurality of data processing engines are configured to perform operations further including: (a) receiving a second plurality of flow cell images from an integrated circuit; (b) determining polymerase clones from the second plurality of flow cell images; and (c) performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images; and (d) sending the second plurality of flow cell images, the determined polymerase clones, and the corresponding base readouts to a first processor or one or more hardware processors or combinations thereof of the sequencing system.

[0157] In some embodiments, one or more operations performed by the first reconfigurable logic device further include sending a second plurality of flow cell images, a determined polymerase clone, a corresponding base readout, or a combination thereof to a first processor or one or more hardware processors of the sequencing system. In some embodiments, one or more operations performed by an integrated circuit further include sending a second plurality of flow cell images, a corresponding base readout, or both to a first reconfigurable logic device, a first processor, or one or more hardware processors of the sequencing system.

[0158] In some embodiments, the operation performed by the first reconfigurable logic device or integrated circuit further includes: registering a second plurality of flow cell images to a common coordinate system.

[0159] In some embodiments, the operation performed by the integrated circuit further includes one or more of the following: determining polymerase clones from a second plurality of flow cell images; performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images; and sending the second plurality of flow cell images, the corresponding base readout, or both to a first reconfigurable device, a first processor, or one or more hardware processors of the sequencing system.

[0160] In some embodiments, the operation (d) or (i) of determining polymerase clones from a second plurality of flow cell images includes: generating a 3D polymerase clone map containing the spatial locations of the polymerase clones based on the determined polymerase clones. The operation of generating the 3D polymerase clone map containing the spatial locations of the polymerase clones based on the determined polymerase clones may further include: deleting duplicate polymerase clones from the determined polymerase clones, wherein the duplicate polymerase clones are out of focus. In some embodiments, the operation of determining polymerase clones from a second plurality of flow cell images includes: overlaying the second plurality of flow cell images with corresponding cell staining images; and generating a polymerase clone map by including only polymerase clones located within cell boundaries in the corresponding cell staining images. Exemplary embodiments of the method for generating a polymerase clone map are disclosed in U.S. Patent Application No. 18 / 078,820 and PCT Application No. PCT / US2023 / 076125, the entire contents of which are incorporated herein by reference.

[0161] In some embodiments, the sequencing methods disclosed herein include the operations described herein. Such operations may include one or more of the following: (a) acquiring sensor data from one or more sensors of the sequencing system via a first reconfigurable logic device of the sequencing system; (b) processing the sensor data via the first reconfigurable logic device to generate a first plurality of flow cell images; (c) predicting a second plurality of flow cell images via the first reconfigurable logic device using a neural network at least partially deployed on the first reconfigurable device and based on the sensor data or the first plurality of flow cell images; (d) determining polymerase clones from the second plurality of flow cell images via the first reconfigurable logic device; (e) performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images via the first reconfigurable logic device; and (f) optionally sending the second plurality of flow cell images, the corresponding base readouts, or both, to one or more processors of the sequencing system via the first reconfigurable logic device.

[0162] In some embodiments, the sequencing methods disclosed herein include the operations described herein. Such operations may include one or more of the following: (a) acquiring sensor data from one or more image sensors of a sequencing system via a first reconfigurable logic device; (b) processing the sensor data via the first reconfigurable logic device to generate a first plurality of flow cell images; (c) transmitting the sensor data, the first plurality of flow cell images, or both, to an integrated circuit via the first reconfigurable logic device; (d) receiving the sensor data, the first plurality of flow cell images, or both, via the integrated circuit and from the first reconfigurable logic device; (e) predicting a second plurality of flow cell images using a neural network based on the sensor data, the first plurality of flow cell images, or both, via the integrated circuit; (f) determining polymerase clones from the second plurality of flow cell images via the integrated circuit; and (g) performing a corresponding base reading on each of the determined polymerase clones based on the second plurality of flow cell images via the integrated circuit.

[0163] In some embodiments, the sequencing methods disclosed herein include the operations described herein. Such operations may include one or more of the following: (a) obtaining sensor data from one or more image sensors of the sequencing system via a first reconfigurable logic device of the sequencing system to generate a first plurality of flow cell images; (b) transmitting the sensor data, the first plurality of flow cell images, or both, to an integrated circuit via the first reconfigurable logic device; (c) receiving the sensor data, the first plurality of flow cell images, or both, from the first reconfigurable logic device via the integrated circuit of the sequencing system; (d) predicting a second plurality of flow cell images based on the sensor data, the first plurality of flow cell images, or both, using a neural network at least partially deployed on the integrated circuit via the integrated circuit; and (e) transmitting the second plurality of flow cell images to the first reconfigurable logic device of the sequencing system or one or more hardware processors via the integrated circuit.

[0164] In some embodiments, the first reconfigurable routing channel includes one or more electronic nodes, and these electronic nodes are programmable. Electronic nodes here may include connection points in a circuit. An electronic node may include a point where two or more circuit elements are connected together. In some embodiments, the first reconfigurable routing channel includes one or more interconnects. Interconnects may include physical wiring connecting transistors and other components on an integrated circuit. In some embodiments, the reconfigurable routing channel includes one or more memory controllers, for example, Figure 5C 5013 in the example. In some embodiments, the first reconfigurable routing channel includes one or more on-chip networks (NoC), for example, Figure 5C 5013 in the example. In some embodiments, the first reconfigurable routing channel may include one or more of the following: a network on-chip (NoC) and a memory controller.

[0165] The first reconfigurable routing channel can be configured to passively transfer data between components of the sequencing system. For example, the reconfigurable routing channel can be configured in a data processing engine (e.g., Figure 5C 5011 in the middle) and memory devices (e.g., Figure 5C The first reconfigurable routing channel can be configured to allow data communication between the first reconfigurable logic device and one or more memory devices. The first reconfigurable routing channel can also be configured to allow data communication between the first reconfigurable logic device and an integrated circuit.

[0166] The reconfigurable logic device described herein may include one or more data processing engines. Each data processing engine may include multiple digital logic circuits.

[0167] The first reconfigurable logic device may be configured to communicate data with one or more external memory devices. The first reconfigurable logic device may be configured to transfer data with one or more external memory devices via a first reconfigurable routing channel. The first reconfigurable logic device may include a first integrated circuit forming an FPGA device. For example, Figure 5C The FPGA device in the FPGA includes a first reconfigurable logic device, a DMA connection, and a first reconfigurable routing channel (e.g., NoC and memory controller).

[0168] The sequencing system may further include one or more memory devices electrically connected to communicate data with one or more components of the sequencing system, the one or more components including one or more of the following: a first reconfigurable logic device; an integrated circuit; a first reconfigurable routing channel; one or more memory controllers; a first processor; a second processor; and one or more processors of the sequencing system.

[0169] In some embodiments, the sequencing system further includes one or more direct data access (DMA) connections, for example, Figure 5C In the 5012, these connections are with multiple data processing engines and the first reconfigurable routing channel (e.g., Figure 5C Data communication is performed between components of the sequencing system (e.g., 5013). DMA connections can be configured to actively communicate data between components of the sequencing system. For example, a DMA connection can be configured to transmit data from a component it is connected to (e.g., 5013). Figure 5C 5011 and Figure 5C(5013) Acquires or sends data to. The DMA connection described herein can be configured to actively request data from or actively send data directly to: a first reconfigurable logic device; a first reconfigurable routing channel; an integrated circuit; or a combination thereof. One or more Direct Data Access (DMA) connections can communicate with the first reconfigurable routing channel and the integrated circuit described herein. The DMA connection can be configured to allow data communication based on a predetermined protocol (e.g., the PCIe protocol).

[0170] In some embodiments, the first reconfigurable routing channel is configured to allow data communication between the first reconfigurable logic device and one or more memory devices. In some embodiments, the one or more DMA connections and the first reconfigurable routing channel are configured to allow data communication between the first reconfigurable logic device and an integrated circuit.

[0171] In some embodiments, the sequencing system further includes an integrated circuit, which is connected to a first reconfigurable logic device (e.g., Figure 5C This differs from 120_b). The integrated circuit described herein may be non-reconfigurable. The integrated circuit may include an application-specific integrated circuit (ASIC) chip. In some embodiments, the integrated circuit includes a neural processing unit (NPU) or an artificial intelligence (AI) chip. The integrated circuit may include a second plurality of data processing engines, each data processing engine including a plurality of digital logic circuits. The integrated circuit may further include: a second plurality of data processing engines and a second routing channel, each routing channel connecting at least some of the second plurality of data processing engines. In some embodiments, the sequencing system further includes a first processor. The first processor may be configured to selectively activate or deactivate different combinations of the first plurality of data processing engines and the first reconfigurable routing channel to perform the operations disclosed herein.

[0172] In some embodiments, a first processor or a second processor is configured to selectively activate or deactivate different combinations of a second plurality of data processing engines and a second reconfigurable routing channel to perform operations. The first processor or the second processor may be configured to selectively activate or deactivate different combinations of a second plurality of data processing engines and a second reconfigurable routing channel to perform the operations described herein.

[0173] The sequencing system may further include a housing that encloses a first reconfigurable logic device, a first reconfigurable routing channel, the one or more DMA connections, an integrated circuit, and a first processor. In some embodiments, the sequencing system further includes a housing that encloses at least the first reconfigurable logic device, and the integrated circuit is located outside the housing.

[0174] In some embodiments, the sequencing system further includes a power supply configured to provide different power levels to the first reconfigurable logic device and the integrated circuit. While sequencing runs and / or sequencing analyses are in progress, a first power level supplied to the first reconfigurable logic device may be higher than a second power level supplied to the integrated circuit. The maximum output power of the sequencing system's power supply is 2, 3, 5, 8, 10, or 20 times lower than the maximum output power of a sequencer (e.g., a conventional sequencer) that does not include the first reconfigurable logic device (e.g., an FPGA), the integrated circuit (e.g., an AI chip), or both. The time consumed by performing a sequencing run and its corresponding sequencing analysis (e.g., a primary analysis) using the sequencing system is 2, 3, 5, 8, 10, or 20 times shorter than the time consumed by performing the same sequencing run using a sequencer that does not include the first reconfigurable logic device, the integrated circuit, or both (e.g., a conventional sequencer that does not include an FPGA and / or AI chip). The time consumed by using a sequencing system to perform a sequencing run and the sequencing analysis of that sequencing run (e.g., primary analysis) is 2, 3, 5, 8, 10, or 20 times shorter than the time consumed by using a sequencer that does not include a first reconfigurable logic device, integrated circuit, or both (e.g., a conventional sequencer that does not include an FPGA and / or AI chip) to perform the same sequencing run and analysis.

[0175] In some embodiments, when performing sequencing runs and corresponding sequencing analyses, the maximum output power supplied to the sequencing system by the power source is less than 900 watts, 800 watts, 700 watts, 650 watts, 600 watts, 550 watts, or 500 watts.

[0176] In some embodiments, the sequencing system further includes a power source configured to supply a first power level to a first reconfigurable logic device, the first power level being less than 500 watts, 400 watts, 350 watts, or 300 watts.

[0177] In some embodiments, the sequencing system further includes a power source configured to supply a second power level to the integrated circuit, the second power level being less than 450 watts, 400 watts, 350 watts, or 300 watts.

[0178] In some embodiments, one or more components are located on a first printed circuit board (PCB). These components may include: a first reconfigurable logic device; a first reconfigurable routing channel; a first processor; and one or more DMA connections. In some embodiments, the integrated circuit is located on a second printed circuit board (PCB) different from the first printed circuit board, for example, as shown below. Figure 5CAs shown in the diagram. The integrated circuit and the second PCB may be located within the same sequencing system housing as the first PCB, or outside the sequencing system housing. Arranging them on separate PCBs facilitates convenient, efficient, and easily customizable interconnection of the first reconfigurable logic device (e.g., an FPGA device) with various integrated circuits on the chip. In some embodiments, the first PCB may be a motherboard, and the second PCB may be a daughterboard.

[0179] In some embodiments, the sequencing system lacks any graphics processing unit (GPU) or tensor processing unit (TPU). Instead, the sequencing system utilizes an FPGA, AI chip, NPU, or other ASIC chip to perform the operations disclosed herein. The sequencing system disclosed herein advantageously requires less power, generates less heat, and reduces the hardware costs for performing NGS sequencing runs and corresponding sequencing analyses.

[0180] A first processor may be located on a first PCB board along with a reconfigurable logic device for convenient and efficient control of the reconfigurable logic device. In some embodiments, the first processor is a separate processor from one or more processors of the sequencing system, and the first processor is configured to control an optical system, fluids of the sequencing system, etc. In some embodiments, the first processor may be configured to control only components on the first PCB board (e.g., an FPGA device), used alone or in combination with components on a second PCB board (e.g., an AI chip). In some embodiments, the sequencing system may include a second processor configured to independently control the AI ​​chip. The first or second processor of the sequencing system (e.g., 120_c) may include a CPU. The one or more hardware processors of the sequencing system include a CPU.

[0181] Continue to refer to Figure 5CIn this particular embodiment, sensor data at imager 116 can be directly transmitted to the data processing engine 5011 of the first reconfigurable logic device 120(a). Alternatively, the sensor data can be stored in a memory device (e.g., 5030) so that the data processing engine can access it. The first processor 120_c controls the operation of the data processing engine and routing channels to process the sensor data and generate a first plurality of flow cell images. Processing may include operations disclosed herein, such as intensity normalization, color correction, hysteresis and advance correction, background subtraction, etc. The first plurality of flow cell images can then be transmitted from the processing engine to the memory device 5030 via the routing channels so that the integrated circuit can be controlled by the first or second processor to access the first plurality of flow cell images for subsequent steps of primary analysis. Alternatively, the first plurality of flow cell images can be directly transmitted to integrated circuit 120-b via DMA connection 5012. In this embodiment, the integrated circuit is used only to predict higher resolution polymerase clone locations using a pre-trained CNN, thereby generating a second plurality of flow cell images with a resolution at least eight times higher than that of the first plurality of flow cell images. CNNs can be pre-trained using simulated images or real flow cell images. A second set of flow cell images is transferred from the integrated circuit back to the first reconfigurable logic device for subsequent processing steps, such as base determination. The base determination results, along with quality information, can then be saved to a FastQ data file. Other information (including cell segmentation and staining) can also be saved in the same file or in another FastQ file with a compatible data format.

[0182] In some embodiments, the sequencing system may further include a heat sink configured to maintain the system temperature in the range of 0°C to 120°C or below 120°C.

[0183] In some embodiments, the operation of processing sensor data to generate a first plurality of flow cell images includes one or more of the following: registering the first plurality of flow cell images to a reference coordinate system; adjusting the image intensity of the first plurality of flow cell images; performing color correction on the first plurality of flow cell images; performing hysteresis and advance correction on the first plurality of flow cell images; and subtracting the background intensity from the first plurality of flow cell images.

[0184] In some embodiments, each of one or more operations performed by a first reconfigurable logic device or integrated circuit is performed within a time window of a single sequencing cycle in which the sequencing run is performed. Figure 5DAn exemplary embodiment of performing sequencing analysis in parallel with a sequencing run is illustrated. In this particular embodiment, the sequencing run includes multiple sequencing cycles. For each cycle, flow cell images are acquired from different color channels at multiple z-levels. Within each cycle, the sequencing reaction is repeated for each z-level within a time window 5601. Operation of the integrated circuit is performed within a processing window 5602 within a single sequencing cycle time window 5609, and also within a time window 5601 for sequencing reactions and imaging at a single z-level 5601. Operation of a first reconfigurable logic device (e.g., onboard primary analysis operation) is also performed within a processing window 5603 within each sequencing cycle time window 5609. Processing windows 5602 and 5603 may be the same or different, depending on various factors such as sequencing data, primary analysis algorithms, etc. In some embodiments, operations are performed not only within the processing window but also within the processing window for data from the current cycle, e.g., for data from the previous z-level for acquired sensor data. In some embodiments, operations are performed within the processing window for data from the previous cycle, e.g., the cycle immediately preceding the current cycle.

[0185] In embodiments where the sample is a 3D volume sample, these operations are performed within a predetermined time window (e.g., 5602, 5603) within a single z-level of each cycle. The predetermined time window is for a single z-level in a single sequencing cycle. In some embodiments, the predetermined time window is less than 1000 ms, 900 ms, 800 ms, 700 ms, 600 ms, 500 ms, 400 ms, 300 ms, 250 ms, 200 ms, or 100 ms. In some embodiments, each of the one or more operations is performed in parallel within the predetermined time window while the sequencing run is in progress. In some embodiments, each of the one or more operations is performed in parallel within a time window after sequencing, imaging, or both are completed in a subsequent sequencing cycle.

[0186] The first plurality of flow cell images described herein may be obtained from multiple z-levels covering at least a portion of the in-situ sample (e.g., cells or tissue). The first plurality of flow cell images may be obtained from one or more color channels of each z-level covering at least a portion of the in-situ sample. In some embodiments, the first plurality of flow cell images are derived from a single color channel. The first plurality of flow cell images may have a first spatial resolution in the x, y, and / or z directions. A second plurality of flow cell images may be generated based on the first plurality of flow cell images. The second plurality of flow cell images may have a second spatial resolution in the x, y, and / or z directions. The first spatial resolution may be lower than the second spatial resolution, and higher resolution herein indicates smaller pixel sizes, resulting in finer spatial detail of the polymerase clones in the flow cell images. The first spatial resolution may be 2, 4, 6, 8, 10, 16, 24, 32, or 48 times lower than the second spatial resolution in the x, y, and / or z directions. The first spatial resolution may be at least 2, 4, 6, 8, 10, 16, 24, 32, or 48 times lower than the second spatial resolution in the x, y, and / or z directions. In some embodiments, the first and second resolutions are 3D resolutions. In some embodiments, the first resolution ranges from 0.1 μm to 5 μm. In some embodiments, the second resolution ranges from 0.01 μm to 2 μm. In some embodiments, the second resolution is at least 4, 6, or 8 times larger than the first resolution in all three dimensions.

[0187] In some embodiments, the sequencing system further includes one or more image sensors configured to receive optical signals generated by a sequencing reaction from a sample fixed on a support. The support may include a glass or plastic substrate. The support may be included in a flow cell device. The one or more image sensors may be configured to generate sensor data based on the optical signals. In some embodiments, the sequencing system further includes: one or more hardware processors; and one or more data storage devices storing instructions executable by the one or more hardware processors to cause the one or more hardware processors to perform the operations disclosed herein. The one or more data storage devices may include one or more memory devices. The one or more memory devices may be accessible by the one or more processors, a first processor, a second processor, a first reconfigurable logic device, or an integrated circuit.

[0188] In some embodiments, the one or more processors are separate from the first processor or the second processor. Operations performed by the one or more processors may include one or more of the following: 1) recording sensor data generated in the sequencing system during one or more flow cycles; 2) optionally processing the recorded sensor data; 3) sending the recorded sensor data or optionally processed data to a first reconfigurable logic device or integrated circuit; 4) receiving output from the first reconfigurable logic device or integrated circuit; and 5) generating sequencing analysis results based on the received output. Operations performed by the one or more processors may include one or more of the following: 1) receiving output from the first reconfigurable logic device or integrated circuit; and 2) generating sequencing analysis results based on the received output.

[0189] In some embodiments, sequencing analysis results include preliminary analysis results. In some embodiments, sequencing analysis results include data files in a predetermined data format. In some embodiments, sequencing analysis results include base interpretation results of nucleotide bases in a sample immobilized on a support. In some embodiments, sequencing analysis results include quality measurement results of the base interpretation results of nucleotide bases in a sample immobilized on a support. In some embodiments, sequencing analysis results include quality scores corresponding to the base interpretation results of nucleotide bases in a sample immobilized on a support.

[0190] In some embodiments, the sequencing system further includes: a sample fixed on a support; and an optical system including: an illumination system; an objective lens; and one or more image sensors. The optical system is configured to emit light toward the sample and collect light signals emitted from the sample, thereby generating a first plurality of flow cell images. The support may be included in a flow cell device.

[0191] In some embodiments, the output data includes base identification results of nucleotide bases in a sample immobilized on a support. In some embodiments, the output data includes the identification of base identification positions in two dimensions. In some embodiments, the output data includes the identification of base identification positions in three dimensions.

[0192] In some embodiments, the operations performed using a neural network via a first reconfigurable routing channel or integrated circuit include one or more of the following: generating quality measurement results of base interpretation; and generating a data output file based on the base interpretation.

[0193] In some embodiments, the neural network includes a convolutional neural network (CNN). In some embodiments, the neural network includes U-Net. In some embodiments, the neural network has been trained using a first reconfigurable logic device or integrated circuit. In some embodiments, the first convolution includes a 3D convolution with a convolutional kernel. In some embodiments, the convolutional kernel has at least four dimensions. In some embodiments, the convolutional kernel is m x m x m x n, where m is an integer ranging from 3 to 30, and n is an integer. In some embodiments, n is an integer ranging from 1 to 16384. In some embodiments, the second convolution in operation (1) includes corresponding numbers n, 2, ..., n in the first repetition, the second repetition, the third repetition, and the fourth repetition, respectively. n、4 n and 8 n filters. In some embodiments, the second convolution in (4) includes the corresponding number of 2 in the last repetition, the penultimate repetition, the penultimate repetition, and the penultimate repetition, respectively. n、2 n、4 n and 8 n filters. In some embodiments, n ranges from 4 to 1024.

[0194] In some embodiments, operation (e) of performing corresponding base readings on each of the identified polymerase clones based on a second plurality of flow cell images includes: performing corresponding base readings on each of the identified polymerase clones based on the second plurality of flow cell images and on a fourth plurality of flow cell images, wherein the fourth plurality of images are predicted using a second neural network based on a third plurality of flow cell images. In some embodiments, the third plurality of flow cell images are acquired from one or more color channels different from a single channel, and wherein the third plurality of flow cell images include a first resolution. In some embodiments, the fourth plurality of flow cell images include a second resolution.

[0195] In some embodiments, the first plurality of flow cell images are derived from one or more color channels. In some embodiments, the first plurality of flow cell images exhibit unbalanced nucleotide diversity. In some embodiments, the first plurality of flow cell images include: unbalanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more flow cycles. In some embodiments, the first plurality of flow cell images include: balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more cycles. In some embodiments, two or more different multiply molecules have different insert sequences. In some embodiments, the different insert sequences correspond to different target RNA molecules or target cDNA molecules. In some embodiments, each position of the identified polymerase clone corresponds to a position of the multiply molecule. In some embodiments, the first plurality of flow cell images include optical signals emitted by nucleotide reagents bound to balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support. In some embodiments, the first plurality of flow cell images include: optical signals emitted by nucleotide reagents that bind to the unbalanced diversity of nucleotide bases A, G, C, and T / U in the polymolecule immobilized on a support, in one or more subsequent cycles. In some embodiments, the unbalanced diversity of nucleotide bases A, G, C, and T / U in the polymolecule includes: (1) the percentage of the number of one or more types of nucleotide bases relative to (2) the total number of bases is less than 20%, 15%, 10%, or 5% in one or more cycles. In some embodiments, the balanced diversity of nucleotide bases A, G, C, and T / U in the plurality of polymolecules includes: (1) the percentage of the number of each type of nucleotide base relative to (2) the total number of bases is greater than 10%, 15%, or 20% in one or more cycles.

[0196] In some embodiments, the cell sample contains a spatial density ranging from 102 2 Up to 10 15 Each mm 2 Overloaded multiply molecules. In some embodiments, the cell sample contains a spatial density ranging from 102 3 Up to 10 10 Each mm 2 Overloaded multiply molecules.

[0197] In some embodiments, the downsampling factor is 2, 4, or 8. In some embodiments, the upsampling factor is 2, 4, or 8. In some embodiments, the downsampling factor is 2, 4, 8, 16, 32, or 64. In some embodiments, the upsampling factor is 2, 4, 8, 16, 32, or 64.

[0198] In some embodiments, one or more of operations (a) through (k) are performed concurrently with a sequencing run. In some embodiments, the first plurality of flow cell images are acquired in sequencing cycles ranging from 1 to 500. In some embodiments, the one or more cycles include a current cycle N. In some embodiments, N ranges from 1 to 500. In some embodiments, the one or more cycles include a single cycle ranging from 1 to 500. In some embodiments, the one or more cycles include a plurality of cycles ranging from 1 to 500. In some embodiments, one or more of operations (e.g., operations (a) through (j)) are performed concurrently with sequencing reactions in cycles following the current cycle N, even if such reactions have not yet been performed or are being performed. In some embodiments, the z-axis is orthogonal to the image plane of the flow cell image.

[0199] In some embodiments, performing a first convolution on the first plurality of flow pool images in one or more dimensions includes: performing a first convolution on the first plurality of flow pool images in 3D, thereby generating a first convolution result. In some embodiments, performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result, includes: performing a second convolution on the first convolution result in 3D, thereby generating a second convolution result.

[0200] In some embodiments, performing a first convolution on the first plurality of flow pool images in one or more dimensions includes: performing a first convolution on the first plurality of flow pool images in 2D, thereby generating a first convolution result. In some embodiments, performing a second convolution on the first convolution result in one or more dimensions, thereby generating a second convolution result, includes: performing a second convolution on the first convolution result in 2D, thereby generating a second convolution result.

[0201] In some embodiments, repeatedly performing the upsampling operation includes: (3) performing upsampling on the fourth convolution result according to the upsampling factor, thereby generating a first upsampling result; (4) concatenating the first upsampling result in the current upsampling repetition with the first downsampling result in the previous downsampling repetition, wherein the first upsampling result and the first downsampling result in the previous downsampling repetition have the same size; and (5) performing a second convolution on the first upsampling result in one or more dimensions, thereby generating a fifth convolution result.

[0202] In some embodiments, different combinations of the first plurality of data processing engines are configured to perform operations further including: (a) receiving a second plurality of flow cell images from an integrated circuit; (b) determining polymerase clones from the second plurality of flow cell images; and (c) performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images; and (d) sending the second plurality of flow cell images, the determined polymerase clones, the corresponding base readouts, or combinations thereof to a first processor or one or more hardware processors of the sequencing system.

[0203] In some embodiments, one or more operations performed by the first reconfigurable logic device further include sending a second plurality of flow cell images, a determined polymerase clone, a corresponding base readout, or a combination thereof to a first processor or one or more hardware processors of the sequencing system. In some embodiments, one or more operations performed by an integrated circuit further include sending a second plurality of flow cell images, a corresponding base readout, or both to a first reconfigurable logic device, a first processor, or one or more hardware processors of the sequencing system.

[0204] In some embodiments, the operation performed by the integrated circuit further includes one or more of the following: determining polymerase clones from a second plurality of flow cell images; performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images; and sending the second plurality of flow cell images, the corresponding base readout, or both to a first reconfigurable device, a first processor, or one or more hardware processors of the sequencing system.

[0205] In some embodiments, the operation performed by the first reconfigurable logic device or integrated circuit further includes: registering a second plurality of flow cell images to a common coordinate system.

[0206] In some embodiments, the operation (d) or (i) of determining polymerase clones from a second plurality of flow cell images includes: generating a 3D polymerase clone map containing the spatial locations of the polymerase clones based on the determined polymerase clones. The operation of generating the 3D polymerase clone map containing the spatial locations of the polymerase clones based on the determined polymerase clones may further include: deleting duplicate polymerase clones from the determined polymerase clones, wherein the duplicate polymerase clones are out of focus. In some embodiments, the operation of determining polymerase clones from a second plurality of flow cell images includes: overlaying the second plurality of flow cell images with corresponding cell staining images; and generating a polymerase clone map by including only polymerase clones located within cell boundaries in the corresponding cell staining images. Exemplary embodiments of the method for generating a 3D polymerase clone map are disclosed in U.S. Patent Application No. 18 / 078,820 and PCT Application No. PCT / US2023 / 076125, the entire contents of which are incorporated herein by reference.

[0207] In some embodiments, the method further includes providing a cell sample carrying multiple RNAs, the RNAs including a first target RNA molecule and a second target RNA molecule. In some embodiments, the method further includes generating multiple cDNA molecules within the cell sample, the multiple cDNA molecules including a first target cDNA molecule corresponding to the first target RNA molecule and a second target cDNA molecule corresponding to the second target RNA molecule. In some embodiments, the method further includes contacting the multiple cDNA molecules in the cell sample with a multiple target-specific locking probe, the target-specific locking probes including at least a first plurality of first target-specific locking probes and a second plurality of second target-specific locking probes.

[0208] In some embodiments, the method further includes contacting the plurality of RNA molecules in the cell sample with a plurality of target-specific lock probes, the target-specific lock probes including at least a first plurality of target-specific lock probes and a second plurality of target-specific lock probes.

[0209] In some embodiments, a single lock probe among the first plurality of first target-specific lock probes comprises: a first terminal region and a second terminal region, wherein the first terminal region selectively hybridizes with a first region of a first target cDNA molecule or a first target RNA molecule, and the second terminal region selectively hybridizes with a second region of a first target cDNA molecule or a first target RNA molecule.

[0210] In some embodiments, contacting the plurality of RNA molecules in a cell sample with a plurality of target-specific lock probes includes: hybridizing a first terminal region and a second terminal region of a first target-specific lock probe with a proximal position on a first target cDNA molecule or a first target RNA molecule to form a circularized first target-specific lock probe having a nick or gap between the hybridized first and second terminal regions. In some embodiments, the first target-specific lock probe includes a first target barcode sequence that corresponds to and uniquely identifies a first target cDNA sequence or a first target RNA sequence. In some embodiments, the first target-specific lock probe includes a first target barcode sequence located adjacent to one of the regions of the first target-specific lock probe, which selectively hybridizes with the first target cDNA molecule or the first target RNA sequence. In some embodiments, the first target-specific lock probe includes at least one universal adaptor sequence. In some embodiments, the first target-specific lock probe includes a universal primer binding site for rolling circle amplification primers or a complementary sequence thereof. In some embodiments, the first target-specific lock probe includes a universal compaction oligonucleotide binding site or a complementary sequence thereof. In some embodiments, the method further includes closing at least a first circularized target-specific lock probe and a second circularized target-specific lock probe by performing an enzymatic reaction, thereby generating at least a first covalently closed circular lock probe and a second covalently closed circular lock probe within a cell sample. In some embodiments, the method further includes performing a rolling circle amplification reaction within a cell sample using the first covalently closed circular lock probe and the second covalently closed circular lock probe as template molecules, thereby generating a plurality of polymer molecules, the plurality of polymer molecules including at least a first polymer molecule corresponding to a first target RNA molecule and a second polymer molecule corresponding to a second target RNA molecule. In some embodiments, the first polymer comprises: a tandem repeat unit of: a first target barcode sequence that uniquely identifies a first target RNA or a first target cDNA sequence, a first insert sequence corresponding to the first target RNA or the first target cDNA, and a first sequencing primer binding site or its complementary sequence. In some embodiments, the first polymer further comprises: a universal binding site for amplifying primers or its complementary sequence, and a universal binding site for compacting oligonucleotides or its complementary sequence. In some embodiments, the second multiply comprises: a tandem repeat unit of the following: a second target barcode sequence that uniquely identifies a second target RNA or second target cDNA sequence; a second insert sequence corresponding to the second target RNA or second target cDNA; and a second sequencing primer binding site or its complementary sequence. In some embodiments, the second multiply further comprises: a universal binding site for amplifying primers or its complementary sequence, and a universal binding site for compacting oligonucleotides or its complementary sequence.

[0211] In some embodiments, one or more cycles of performing a sequencing reaction include: contacting a plurality of multiply molecules within a cell sample with (i) a plurality of universal sequencing primers, (ii) a plurality of sequencing polymerases, and (iii) a plurality of nucleotide reagents under conditions suitable for hybridizing the plurality of universal sequencing primers to their respective universal sequencing primer binding sites on the multiply. In some embodiments, the plurality of nucleotide reagents includes: multivalent molecules, nucleotides, nucleotide analogs, or combinations thereof. In some embodiments, a single nucleotide or nucleotide analog is detectably labeled or unlabeled. In some embodiments, a detectably labeled single nucleotide or nucleotide analog contains a different detectable color marker corresponding to each different type of nucleotide base A, G, C, and T / U. In some embodiments, a single multivalent molecule contains a core with a plurality of nucleotide arms attached, and each arm of the single multivalent molecule contains the same type of nucleotide base.

[0212] In some embodiments, generating the first plurality of flow cell images includes: in each cycle, imaging optical color signals emitted from nucleotide reagents bound to the plurality of polyp molecules using an optical system. In some embodiments, the first plurality of flow cell images include optical color signals emitted from nucleotide reagents bound to the plurality of polyp molecules. In some embodiments, one or more cycles of performing a sequencing reaction include: sequencing only the first target barcode sequence region of the first polyp, thereby generating a first sequencing read product. In some embodiments, one or more cycles of performing a sequencing reaction include: sequencing at least a portion of the first target barcode sequence region of the first polyp and a first insert sequence, thereby generating a first sequencing read product.

[0213] In some embodiments, one or more cycles of performing a sequencing reaction include sequencing only the second target barcode sequence region of the second multipartite, thereby generating a second sequencing read product. In some embodiments, one or more cycles of performing a sequencing reaction include sequencing at least a portion of the second target barcode sequence region of the second multipartite and the second insert sequence, thereby generating a second sequencing read product.

[0214] In some embodiments, the method further includes: removing a first sequencing read product from a first polyparticle and retaining the first polyparticle in the cell sample, and removing a second sequencing read product from a second polyparticle and retaining the second polyparticle in the cell sample. In some embodiments, the method further includes: repeatedly sequencing the plurality of polyparticles by repeating the following operations at least once: generating a first plurality of flow cell images of a cell sample immobilized on a support by performing one or more cycles of sequencing reactions, thereby generating a first sequencing read product and a second sequencing product, wherein the cell sample contains a plurality of polyparticles, wherein a first polyparticle among the plurality of polyparticles corresponds to a first target RNA molecule of the cell sample, and a second polyparticle among the plurality of polyparticles corresponds to a second target RNA molecule of the cell sample, wherein the first plurality of flow cell images; and removing the first sequencing read product from the first polyparticle and retaining the first polyparticle in the cell sample, and removing the second sequencing read product from the second polyparticle and retaining the second polyparticle in the cell sample.

[0215] In some embodiments, the first sequencing read product comprises some or all of the following: a first target barcode sequence in one or more tandem units of a first multiply molecule; a first insert sequence in one or more tandem units of a first multiply molecule; or a combination thereof.

[0216] In some embodiments, the method further includes: confirming the presence of a first target RNA molecule, a second target RNA molecule, or both molecules in a cell sample based on base readings performed at base reading positions in a base reading template.

[0217] In some embodiments, the method further includes: generating a second plurality of flow cell images of a cell sample immobilized on a support by performing subsequent cycles of sequencing reactions after one or more cycles using a sequencing system. In some embodiments, generating a first plurality of flow cell images of a cell sample immobilized on a support includes: sequencing at least a first polyp within the cell sample while suppressing sequencing of a second polyp. In some embodiments, sequencing at least a first polyp within the cell sample includes: generating a plurality of first sequencing read products, wherein the sequences of the first sequencing read products are aligned with a first target reference sequence to confirm the presence of a first target RNA in the cell sample. In some embodiments, generating a first plurality of flow cell images of a cell sample immobilized on a support includes: sequencing at least a second polyp within the cell sample while suppressing sequencing of the first polyp. In some embodiments, sequencing at least a second polyp within the cell sample includes: generating a plurality of second sequencing read products, wherein the sequences of the second sequencing read products are aligned with a second target reference sequence to confirm the presence of a second target RNA in the cell sample.

[0218] Predicting high-resolution flow cell images

[0219] Figure 5A A flowchart of a computer-implemented method 500 is shown, which is used to predict high-resolution flow cell images, thereby improving the density of detectable polymerase clones in the flow cell images. Method 500 may include some or all of the operations disclosed herein. These operations may be performed in, but are not limited to, the order described herein.

[0220] Method 500 can be executed by one or more processors disclosed herein. In some embodiments, the processor may include one or more of the following: a computing system including processing unit 118, reconfigurable logic device 120, non-reconfigurable integrated circuit 120, or a combination thereof. For example, the processing unit may include a central processing unit (CPU). The reconfigurable logic device may include one or more FPGA devices. The integrated circuit may include a chip, such as an AI chip or an ASIC chip. In some embodiments, the one or more processors may include the computer system 400 disclosed herein.

[0221] In some embodiments, some or all of the operations in methods 500, 600, 700, 2800, and 2900 may be performed by a reconfigurable logic device (e.g., an FPGA) and / or an integrated circuit (e.g., an AI chip). In embodiments, when some operations are performed by a reconfigurable logic device and / or an integrated circuit (e.g., an FPGA), data generated by the reconfigurable logic device and / or integrated circuit (e.g., an FPGA) after performing one or more operations may be transferred to various hardware elements of system 100 (e.g., a CPU or GPU), allowing subsequent operations in methods 500, 600, 700, 2800, and 2900 to be performed using the transferred data via such hardware. Similarly, data may also be transferred in reverse from various hardware (e.g., a CPU) to the reconfigurable logic device or integrated circuit for processing. In some embodiments, all operations in methods 500, 600, 700, 2800, and 2900 may be performed by a CPU. Alternatively, operations performed by a CPU may be performed by other processors, such as a dedicated processor, or a GPU. In some embodiments, all operations in methods 500, 600, 700, 2800, and 2900 may be performed by reconfigurable logic devices and / or integrated circuits (e.g., FPGAs and / or AI chips).

[0222] In some embodiments, sensor data acquired by imager 116 can be directly transmitted to a reconfigurable logic device and / or integrated circuit (e.g., via a DMA connection). In some embodiments, sensor data acquired by imager 116 can be directly transmitted to a reconfigurable logic device and / or integrated circuit without first being routed to a CPU, GPU, or any other processing unit before reaching the reconfigurable logic device and / or integrated circuit.

[0223] In some embodiments, the power required to predict high-resolution flow pool images using the method 500 described herein and reconfigurable logic devices (e.g., FPGAs) and / or other integrated circuits (e.g., AI chips) may be at least 2, 8, 10, 15, 20, 40, 50, or 100 times lower than the power required to make the same prediction using other computing hardware (including, but not limited to, CPUs or GPUs).

[0224] In some embodiments, the sequencing system of this invention further includes a power supply configured to provide the same or different power levels to the reconfigurable logic devices and integrated circuits. In some embodiments, when performing methods 500, 600, 700, 2800 and / or 2900, the maximum power output provided by the power supply to the sequencing system is less than 2000 watts, 1000 watts, 900 watts, 800 watts, 700 watts, 650 watts, 600 watts, 550 watts, 500 watts, 400 watts, 300 watts, 200 watts, or 100 watts.

[0225] In some embodiments, the sequencing system of this invention includes: a first reconfigurable logic device, such as an FPGA unit, including a plurality of data processing engines configured to perform data processing in parallel; a first reconfigurable routing channel, each connected to at least some of the first plurality of data processing engines; a neural network, at least partially deployed on the first reconfigurable logic device; and a first processor for selectively activating or deactivating different combinations of the first plurality of data processing engines and the first reconfigurable routing channel to perform one or more operations of the methods of this invention (e.g., methods 500, 2800) to make predictions.

[0226] In some embodiments, the sequencing system of this invention includes: a first reconfigurable logic device including a first plurality of data processing engines arranged in a first pipeline and configured to perform data processing in parallel with each other; an integrated circuit that communicates data with the first reconfigurable logic device; a neural network at least partially deployed on the integrated circuit and / or the first reconfigurable logic device; and a first processor of the first reconfigurable logic device for selectively activating or deactivating different combinations of the first plurality of data processing engines to perform one or more operations of the methods of this invention (e.g., methods 500, 2800) to make predictions using the neural network.

[0227] In some embodiments, the first reconfigurable logic device and integrated circuit are housed within the same physical enclosure as other components of the sequencing system, such as... Figure 1 As shown in the diagram. In some embodiments, the first reconfigurable logic device and integrated circuit are not physically outside the sequencing system 110, such as... Figure 1 As shown, for example, not in cloud 130.

[0228] Method 500 may include the following operation 510: (i) generating first plurality of flow cell images of a sample immobilized on a support by performing one or more cycles of sequencing reactions using a sequencing system 110. The sample may contain polymolecules therein. The sample may contain polymolecules from one or more different sample sources. The sample may have a thickness along the z-axis such that first plurality of flow cell images can be acquired at z-stacks at different z-positions with a first resolution to cover the sample in 3D. The sample may be acquired from a single z-position of a 2D or 3D sample.

[0229] The sample can be in situ. The sample can be a 3D sample. The sample can be a volumetric sample, which may contain different biological information at the same xy location but different z-levels. The sample can be a cellular sample comprising multiple cells, tissues, or combinations thereof. The sample can be any biological sample having a thickness along the z-axis greater than a predetermined threshold. For example, the thickness can be greater than 2 μm, 3 μm, 4 μm, 5 μm, 10 μm, 20 μm, or greater. The z-axis (e.g., the z-axis line) is orthogonal to the image plane defined by the x-axis and y-axis. In some embodiments, the sample can be a conventional 2D sequencing sample.

[0230] In some embodiments, such a computer-implemented method includes the following operation: (i) generating, via a sequencing system, one or more cycles of performing a sequencing reaction, a first plurality of flow cell images of a sample immobilized on a support, wherein the first plurality of flow cell images are acquired at a first resolution. This operation is similar to... Figure 5A Operation 510 in the example is applicable, but the sample can be a 2D or 3D sample. In an embodiment where the sample is 3D, the sample contains multiply molecules therein. In an embodiment where the sample is 2D, the sample contains template molecules therein.

[0231] Flow cell images can be acquired from 1, 2, 3, 4 or more color channels using the optical system of the imager 116 disclosed herein. Each flow cell image may include at least a portion of one or more blocks (e.g., imaging regions). Each block may be divided into multiple sub-blocks. Each block or sub-block may include multiple polymerase clones or clusters. Each sub-block may include multiple regions, wherein each region includes several polymerase clones or clusters. Flow cell images as disclosed herein can be acquired from, for example,... Figure 1 The flow cell 112 shown is or Figure 27 The image obtained by 2712 shown is illustrated. In some embodiments, the flow pool image is obtained from a single color channel, and subsequent prediction is performed using a pre-trained neural network corresponding to that single channel. In some embodiments, the flow pool image is obtained from 2, 3, 4, or more color channels, and subsequent prediction is performed using a pre-trained neural network corresponding to those multiple color channels.

[0232] In some embodiments, the flow cell image herein may be an image of one or more blocks, one or more sub-blocks, one or more segmented regions within a block or sub-block, or a combination thereof. Each flow cell image may include a field of view (FOV). The FOV may be perpendicular to the z-axis. The FOV may lie in the xy-plane. The FOVs of different flow cell images at different z-levels may be the same in the xy-plane. The FOVs of different flow cell images at different z-levels may have at least an overlap in the xy-plane. The image resolutions of different flow cell images at different z-levels may be substantially the same or identical. In some embodiments, the image resolutions of different flow cell images at different z-levels are different. Figure 3A and Figure 3D Two exemplary flow cell images acquired at two different z-levels along the z-axis of the same 3D sample within the same sequencing cycle are shown. The field of view (FOV) can be 3D and can have various sizes to cover the volume of sample to be imaged. The FOV can range from 10 μm to 5 mm along the x, y, and / or z directions. The FOV can range from about 0.1 μm to about 2 mm along the x, y, and / or z directions. The FOV can range from 0.5 μm to 1 mm along the x, y, and / or z directions. For example, for some cell samples, the FOV can be about 0.5 mm x 0.5 mm x 20 μm along the x, y, and z directions, respectively.

[0233] The flow cell images described herein can have various sizes, and the number of pixels along the x-axis, y-axis, and / or z-axis can be any integer greater than 64 or 128. The number of pixels along the x-axis, y-axis, and / or z-axis can range from 2 to 65,536. A single flow cell image can be divided into different numbers of regions, such as 4, 8, 16, or even more regions, each region's size can be 256 x 256 x 1, 512 x 512 x 3, or other sizes. In some embodiments, the number of pixels along the x-direction, y-direction, and / or z-direction can be adjusted to maintain a specific spatial resolution within a given FOV. For example, for a spatial resolution of 0.2 μm, the number of pixels could be 4000 to cover an FOV of 0.8 mm.

[0234] Each flow cell image at a specific z-level can contain intensities generated by polymerase clones or clusters at the corresponding z-level. For example... Figure 3A and Figure 3DAs shown, the signals from polymerase clones or clusters are small bright spots within the image. Each bright spot can be of various sizes, less than a few pixels, for example, less than one pixel, about one pixel, about two pixels, three pixels, four pixels, five pixels, or larger. In some embodiments, each signal point of a polymerase clone or cluster can be any number of pixels ranging from 0.01 pixels to about 100 pixels. In some embodiments, each signal point of a polymerase clone or cluster can be any number of pixels ranging from 0.1 pixels to about 16 pixels.

[0235] Each flow cell image may also include intensities generated by cells and their structural elements. Such structural elements can be background objects or components, for example, in... Figure 3A In addition, each flow cell image may also include noise and / or artifacts that are not derived from polymerase clones or cell structures.

[0236] In some embodiments, when the depth of field of the optical system includes a range extending along the z-axis (e.g., 0.1 μm, 0.2 μm, 0.3 μm, 0.5 μm, 0.6 μm, 0.8 μm, 1 μm, 2 μm, 3 μm, 4 μm, 5 μm, etc.), polymerase clones or clusters within this depth of field may appear in focus or nearly in focus in the flow cell image. Flow cell images at specific z-levels may also include signals from polymerase clones or clusters that are not within the image's focus range. Therefore, such polymerase clones or clusters are out of focus. Figure 3A As shown, larger and blurry signal points represent out-of-focus polymerase clones or clusters. Figure 3A Some of the defocused polymerase clones or clusters are circled in the middle.

[0237] Each flow cell image at a specific z-level may also include noise caused by the optical system and / or unwanted signals from the sample. Unwanted signals can be signals from sample components such as cell membranes, cytoplasm, and mitochondria. Such background objects can be any object that is relatively large in size compared to a polymerase clone or cluster. Figure 3A As shown, the flow cell image contains blurred cell outlines (indicated by arrows), and most signal points are contained within these blurred outlines. In some embodiments, background objects may include any object that is within the 3D sample but is not a polymerase clone or cluster.

[0238] In some embodiments, the flow cell image is derived from multiple color channels. In some embodiments, the flow cell image exhibits unbalanced nucleotide diversity. In some embodiments, the flow cell image includes: unbalanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more sequencing cycles. In some embodiments, the flow cell image includes: balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more cycles. In some embodiments, two or more different multiply molecules have different insert sequences. In some embodiments, the different insert sequences correspond to different target RNA molecules or target cDNA molecules. In some embodiments, each position of the identified polymerase clone corresponds to a position of the multiply molecule. In some embodiments, the flow cell image includes optical signals emitted by nucleotide reagents that bind to the balanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support. In some embodiments, the flow cell image includes: optical signals emitted by nucleotide reagents that bind to the unbalanced diversity of nucleotide bases A, G, C, and T / U in multiply molecules immobilized on a support during one or more subsequent cycles. In some embodiments, unbalanced diversity of nucleotide bases A, G, C, and T / U in a polyp molecule includes: (1) the percentage of the number of one or more types of nucleotide bases relative to (2) the total number of bases is less than 20%, 15%, 10%, or 5% in one or more sequencing cycles. In some embodiments, balanced diversity of nucleotide bases A, G, C, and T / U in the polyp molecule includes: (1) the percentage of the number of each type of nucleotide base relative to (2) the total number of bases is greater than 10%, 15%, or 20% in one or more cycles. As an example, base interpretation results from polymerase clones include four different bases, and for each of the four bases, the percentage of the polymerase clone may be greater than about 10%, thus the data has balanced diversity. As another example, base interpretation results from the polyp molecule include four or fewer different bases, and for one or more bases, the percentage of the polymerase clone may be less than about 10%, and such data can be considered unbalanced diversity. In some embodiments, the base interpretation results from the plurality of polymerase clones include four or fewer different bases, and for some of the bases, the percentage of the polymerase clones may be less than about 5%, about 2%, or even about 1%, and such data can be considered as unbalanced diversity. As another example, the unbalanced diversity data in the plurality of polymerase clones includes bases A, T, C, and G, and they account for about 1%, about 2%, about 1%, and about 95% of the total base interpretation results, respectively.Besides base bias affecting diversity, complexity is also a factor; when complexity is below a certain number (e.g., 8 or 16), the signal exhibits unbalanced diversity.

[0239] Method 500 is configured to predict high-resolution flow cell images even if polymerase clones in the acquired flow cell images have uneven diversity in one or more sequencing cycles.

[0240] In some embodiments, method 500 includes the following operation 520: (ii) providing a first plurality of flow cell images as input to a neural network (e.g., a CNN) via a processor, wherein the neural network is pre-trained using the training method 600 of this invention and a training dataset of training flow cell images. The neural network is pre-trained such that the parameter values ​​of the neural network have been optimized based on training. The neural network may be retrained if necessary, for example, for predicting flow cell images from different cell samples.

[0241] In some embodiments, the computer-implemented method 500 can be used to predict high-resolution flow cell images with a resolution higher than that of first plurality of flow cell images acquired by imager 116 (e.g., 2x, 4x, or 6x along at least one spatial dimension). In some embodiments, the high-resolution flow cell image can be a post-processed image of the first plurality of flow cell images, for example, through... Figure 31 The image processing part 3120 of the neural network performs processing. The image processing in this paper may include various image processing steps, including but not limited to: background removal, background reduction, artifact removal, artifact suppression, signal-to-noise ratio adjustment, contrast-to-noise ratio adjustment, intensity normalization, intensity shift correction, noise reduction, color correction, hysteresis or loss of synchronization correction, image registration, and deconvolution.

[0242] In some embodiments, the neural network in operation 520 is a first neural network that can be trained using the method 700 disclosed herein.

[0243] In some embodiments, method 500 includes operation 520', which replaces operation 520. In some embodiments, operation 520 includes (ii) providing a first plurality of flow cell images as input to a neural network via a processor or a first reconfigurable logic device, wherein the neural network is pre-trained using a training dataset of training flow cell images and reference base interpretation results of the training dataset.

[0244] In some embodiments, operation 520' is similar to operation 520, for example, as Figure 5AAs shown, the difference lies in the different neural networks. In some embodiments, operation 520' can replace operation 520 in method 500. In some embodiments, the neural network in operation 520' is a different neural network from the neural network in operation 520.

[0245] In some embodiments, the neural network in operation 520 is a first neural network, and the neural network in operation 520' is a second neural network different from the first neural network in operation 520. The differences between the first neural network and the second neural network may include, but are not limited to: different types of neural networks, differences in parameter values, the number of parameters, the number of convolutional layers, the number of layers, or combinations thereof.

[0246] In some embodiments, the second neural network in operation 520' is a different neural network pre-trained using the same flow pool image training dataset as the flow pool image training dataset used to train the first neural network in operation 520.

[0247] In some embodiments, the second neural network in operation 520' is a different neural network pre-trained using a flow pool image training dataset different from the flow pool image training dataset used to train the first neural network in operation 520. In some embodiments, the second neural network in operation 520' is pre-trained using reference base interpretation results from the training dataset.

[0248] In some embodiments, the first neural network in operation 520 is pre-trained using a reference image or reference intensity as a benchmark ground truth (e.g., a reference high-resolution image or a reference intensity in a high-resolution image), and the second neural network in operation 520 is pre-trained using the reference base interpretation results of the training flow cell images in the training dataset as benchmark ground truth.

[0249] In some embodiments, the reference base reading result may be generated using various base reading methods, including those methods described herein that relate to the training methods used to predict the base reading result herein. In some embodiments, the reference base reading result may be generated using methods lacking neural networks. Exemplary embodiments for generating base reading results from flow cell images are disclosed in U.S. Patent Application No. 18 / 078,820 and PCT Application No. PCT / US2023 / 076125, all of which are incorporated herein by reference.

[0250] In some embodiments, the second neural network in operation 520' can use a similar... Figure 5EThe training method of method 700 is used for training, in which reference strength is not used for operations, such as operations 725, 730, and 755. Instead, reference base interpretation results are used in such operations (e.g., operations 725', 730', and 755).

[0251] In some embodiments, the loss function used to train the second neural network in operation 520' may differ from the loss function used to train the first neural network in operation 520. In some embodiments, various loss functions may be used to train the second neural network in operation 520'. In some embodiments, the second neural network is pre-trained using one or more loss functions based on comparing training base interpretation results of training flow cell images with reference base interpretation results of training flow cell images. In some embodiments, the loss function used to train the second neural network in operation 520' may be based on comparing training outputs (e.g., base interpretation results) with reference base interpretation results. In some embodiments, training of the second neural network in operation 520' is complete when the loss function satisfies a predetermined criterion. The predetermined criterion may be customized to include various aspects of the training output. In some embodiments, the predetermined criterion is determined based on a comparison of training base interpretation results with reference base interpretation results. In some embodiments, the predetermined criterion is based on the correctness of the comparison between training base interpretation results and reference base interpretation results. In some embodiments, the predetermined criterion is based at least on the training time spent.

[0252] Figure 31 The diagram illustrates exemplary embodiments of a first neural network and a second neural network, as well as a method for training such neural networks.

[0253] It is worth noting that although a neural network 3110 is disclosed in this exemplary embodiment, the neural network 3110 herein can be any artificial intelligence-based or machine learning-based model, which may include an image processing section 3120 and a base determination section 3130. Similarly, the image processing section 3120 and the base determination section 3130 can each be any artificial intelligence-based or machine learning-based model that can achieve similar functionality to an equivalent neural network-based model.

[0254] In some embodiments, neural network 3110 may be a first neural network in operation 520 or a second neural network in operation 520'. As an example, the method for training neural network 3110 may be method 700. The neural network may include two separate parts, a first part being an image processing part 3120 and a second part being a base determination part 3130.

[0255] Image processing unit 3120 is configured to perform one or more image processing steps disclosed herein on the flow cell images (e.g., a first plurality of flow cell images or a second plurality of flow cell images) herein, such as steps related to method 500. These one or more image processing steps may include, but are not limited to: background removal, background reduction, artifact removal, artifact suppression, signal-to-noise ratio adjustment, contrast-to-noise ratio adjustment, intensity normalization, intensity shift correction, noise reduction, color correction, hysteresis or out-of-synchronization correction, image registration, intensity extraction, and deconvolution.

[0256] In some embodiments, the base interpretation section 3130 is configured to perform base interpretation using the output image 3150 from the image processing section 3120 of the neural network. The base interpretation section 3130 may be configured to perform such image processing steps in embodiments where intensity extraction, color correction, and / or hysteresis or out-of-synchronization correction are not performed in the image processing section 3120 of the neural network.

[0257] The first portion 3120 or the second portion 3130 of the neural network may each include one or more neural network structural elements, such as convolutional layers. In some embodiments, the first portion 3120 or the second portion 3130 of the neural network may include one or more embedding layers of the neural network. The first portion 3120 of the neural network may include at least a portion of the encoder of the neural network, and the second portion of the neural network may include at least a portion of the decoder of the neural network. In some embodiments, the second portion of the neural network may include at least a portion of the following: convolutional layers, pooling layers, fully connected layers, SoftMax layers, input layers, output layers, embedding layers, encoders, and decoders of the neural network.

[0258] In some embodiments, the base-reading section 3130 may lack any structural elements of the neural network, such as convolutional or pooling layers. In some embodiments, the base-reading section 3130 may lack any artificial intelligence-based algorithm. In some embodiments, the base-reading section 3130 may lack any convolutional layers of the neural network. In some embodiments, the base-reading section 3130 may lack any part of the embedding layer or decoder of the neural network. In some embodiments, the base-reading section 3130 may lack any of the following: convolutional layers, pooling layers, fully connected layers, SoftMax layers, input layers, output layers, embedding layers, encoders, and decoders of the neural network. In some embodiments, the base-reading section 3130 may include only non-neural network base-reading algorithms. As an example, the neural network 3110 that generates the output image 3150 is the second neural network in operation 520'. As another example, the neural network 3110 that generates the output base-reading result 3160 is the neural network disclosed with respect to method 2800.

[0259] In embodiments where the base-reading section 3130 lacks any element of a neural network or artificial intelligence-based algorithm, training the neural network may include training one or more parameters of the base-reading section 3130. For example, these one or more parameters may include feature sizes. In such embodiments, during training, backpropagation (e.g., derived from a loss function, reference values, and output base-reading results) for finding adjustments to the parameter values ​​of the neural network 3110 passes through the base-reading section 3130 and makes any adjustments (parameter adjustments) to the base-reading section 3130 and the image processing section 3120, such as... Figure 31 The solid gray line with an arrowhead is shown.

[0260] In some embodiments where the base interpretation section 3130 lacks any element of a neural network or artificial intelligence-based algorithm, the training of the neural network does not include training any parameters of the base interpretation section 3130. During training, backpropagation (e.g., derived from the loss function, reference values, and output base interpretation results) used to find adjustments to the parameter values ​​of the neural network 3110 may pass through the base interpretation section 3130 without making any adjustments to the base interpretation section 3130 and subsequently the image processing section 3120 (but parameter adjustments are performed), such as... Figure 31 The solid gray line with an arrowhead is shown.

[0261] In some embodiments, when the base determination section 3130 includes at least some elements of a neural network or an artificial intelligence-based algorithm, training the neural network may include training the base determination section 3130 and the image processing section 3120, including adjusting parameters from these two sections, such as... Figure 31 The gray solid line with an arrow in the middle indicates this.

[0262] In other embodiments where the base-reading section 3130 lacks any element of a neural network or artificial intelligence-based algorithm, the training of the neural network may consist only of the training of the image processing section 3120 but not of the training of any of the parameters in the base-reading section 3130. In such embodiments, during training, backpropagation (e.g., from the loss function, reference values, and output base-reading results) for updating the parameters of the thus trained neural network 3110 goes directly to the image processing section 3120 without passing through the base-reading section 3130, such as... Figure 31 The gray dashed line with arrows indicates this. In other words, in this type of embodiment, the base interpretation section 3130 is untrained and the parameters in the base interpretation section 3130 are fixed. In this type of embodiment, the loss function may be based on the output of the base interpretation section 3130, and the value of the loss function may be determined based on the output of the base interpretation section.

[0263] After training the neural network, input 3140 may pass through image processing section 3120 to generate output image 3150. In some embodiments, output image 3150 includes a second plurality of flow cell images, such as those disclosed herein with respect to method 500. In some embodiments, output image 3150 includes a high-resolution post-processed image corresponding to input image 3140. Output image may pass through base determination section 3130 to generate base determination result 3160. Alternatively, output image 3150 may pass through various base determination algorithms (e.g., conventional base determination algorithms not based on neural networks) but not through base determination section 3130 used to generate base determination result. In such embodiments, after training the model, only image processing section 3120 of the neural network is used to make predictions, without using base determination section 3130 of the neural network. In such embodiments, neural network 3110 advantageously reduces the time required to make predictions and reduces the computational burden and power required to make predictions compared to existing neural networks that predict base determination results.

[0264] In some embodiments, input image 3140 includes a raw flow cell image acquired at imager 116. In some embodiments, input image 3140 includes a first plurality of flow cell images disclosed herein. In some embodiments, input image 3140 may be derived from a plurality of color channels and a plurality of sequencing cycles. In some embodiments, input image 3140 may be derived from a plurality of color channels and a single sequencing cycle. In some embodiments, input image 3140 may be derived from a single color channel and a plurality of sequencing cycles. In some embodiments, input image 3140 may be derived from a single z-level or multiple z-levels.

[0265] See also Figure 31In embodiments where neural network 3110 is trained (e.g., using method 700), a reference value or baseline true value 3180 can be used to compare the output base reading result 3160, and a loss function 3170 is calculated based on this comparison. The value of the loss function can then be used during training to backpropagate into neural network 3110 to adjust the parameter values ​​of neural network 3110, such as gradients. In some embodiments, adjusting the parameters of the neural network may include the parameters of the base reading portion 3130 and the image processing portion 3120. In other words, portions 3120 and 3130 are trained during neural network training (e.g., using training methods 700, 2900 herein). In some embodiments, the value of the loss function can be used to backpropagate into neural network 3110 to adjust only the parameter values ​​of the image processing portion 3120, without adjusting the parameter values ​​of the base reading portion 3130. In other words, in such embodiments, only the image processing portion 3120 is trained during the training of the neural network (e.g., using the training methods 700, 2900 herein), while the base determination portion 3130 is not trained. In some embodiments, the neural network 3110 trained only on the image processing portion 3120 is a second neural network in operation 520'. In some embodiments, the neural network 3110 trained on both the image processing portion 3120 and the base determination portion 3130 is a second neural network in operation 520'.

[0266] In some embodiments, the second neural network in operation 520' includes a convolutional neural network. In some embodiments, the second neural network in operation 520' includes a recurrent neural network. In some embodiments, the second neural network in operation 520' includes a U-Net, residual U-Net, ResNet (residual neural network), and / or LSTM (long short-term memory) neural network.

[0267] In some embodiments, training flow cell images are acquired only from the same color channel. In some embodiments, each of the training flow cell images includes flow cell images from the same field of view from multiple sequencing cycles stacked along the time dimension. The multiple sequencing cycles may be in the same sequencing run. The multiple sequencing cycles may be consecutive sequencing cycles in a sequencing run. In some embodiments, each of the training flow cell images includes flow cell images from the same field of view from one or more sequencing cycles. In some embodiments, each of the training flow cell images includes flow cell images of samples at one or more z-levels. In some embodiments, training flow cell images include flow cell images of the same sample at multiple different fields of view. In some embodiments, training flow cell images include flow cell images of one or more samples at multiple different fields of view. Different fields of view may be at the same x, y, or z position of the same sample. For example, different fields of view may be different sub-blocks of a sample at the same z position but different x, y positions. For example, multiple different fields of view may be adjacent to each other and have no or at least some spatial overlap with other fields of view. In some embodiments, each of the training flow cell images includes flow cell images from different fields of view (e.g., adjacent FOVs of the same sample) from multiple sequencing cycles stacked along one or two spatial dimensions.

[0268] In some embodiments, the training dataset for the neural network (e.g., a first or second neural network) may consist only of flow cell images of the same color channel. Thus, the neural network is not trained to adapt to variations across different color channels, which could be caused by differences in optical elements responding to different colors of light signals (e.g., emission filters, illumination, etc.), differences in fluorescent dyes, or other factors of the sequencing system. In some embodiments, such variations can lead to, but are not limited to, different background levels, different signal-to-noise ratios, different artifacts in the field of view, different full width at half maximum (FWHM) of the emitted light signal, different point spread functions (PSF), etc. Training the neural network using flow cell images of the same color channel advantageously eliminates fitting to variations across different color channels and simplifies and accelerates the training of the neural network while avoiding potential errors during training.

[0269] In some embodiments, different neural networks are trained using flow cell images corresponding to the color channels. In other words, the neural networks are trained as channel-specific neural networks. In an embodiment using three colors for sequencing, three different neural networks are trained using corresponding flow cell images for the corresponding color channels. Each channel-specific neural network is used to predict the high-resolution flow cell image for the corresponding color channel.

[0270] In embodiments where two or more channels have the same color but different wavelength ranges, a single neural network can be trained using flow cell images of the same color from two or more channels. Such a neural network can be used to predict high-resolution flow cell images or to make inferences about high-resolution flow cell images from two or more channels of the same color.

[0271] In some embodiments, when two or more channels have the same color but different wavelength ranges, a single-channel flow cell image can be used to train different neural networks. Each different neural network is a channel-specific neural network that can be used to make predictions or inferences only for the corresponding channel.

[0272] In some embodiments, after operation 520', method 500 includes the following operation 530: (iii) using a neural network via a first reconfigurable device or integrated circuit to predict a second plurality of flow cell images, wherein each of the second plurality of flow cell images has a second resolution and corresponds to a corresponding image of the first plurality of flow cell images, and wherein the second resolution is at least 2 to 32 times larger than the first resolution in one or more spatial dimensions.

[0273] In some embodiments, (iii) the operation 530 of using a neural network to predict a second plurality of flow cell images includes predicting a high-resolution post-processed image corresponding to the first plurality of flow cell images, wherein the processing includes various image processing or intensity processing steps. For example, the processing steps may include one or more of the following: noise reduction, background reduction; background removal; artifact removal; artifact suppression; intensity shift correction; intensity normalization; signal-to-noise ratio adjustment; contrast-to-noise ratio adjustment; color correction; hysteresis and / or out-of-sync; image registration; and deconvolution. In some embodiments, predicting a high-resolution post-processed image corresponding to the first plurality of flow cell images may advantageously allow the generation of a higher resolution and higher image quality version of the first plurality of flow cell images, and this higher resolution and higher quality version may be used to generate more accurate and reliable base interpretation results. To achieve more accurate and reliable base interpretation, the second neural network of method 500, for example in operation 520', may be trained without using a reference flow cell image or reference intensity, but using a reference base interpretation result as a baseline ground truth. Reference base interpretation results can be generated using various methods, including those disclosed herein regarding methods for training a neural network to predict base interpretation results. In some embodiments, for example, training using method 700 or other training methods herein can optimize at least some of the neural network parameters to produce training base interpretation results that are sufficiently similar to the reference base interpretation results (e.g., determined by the value of a loss function that satisfies predetermined criteria). As a result, the trained neural network can be used to predict high-resolution post-processed images corresponding to a first plurality of flow cell images, and such high-resolution post-processed images can be used to produce accurate and reliable base interpretation results. Compared to embodiments of method 500 with operation 520 for predicting high-resolution flow cell images or other existing methods for directly predicting base interpretation results, embodiments of method 500 with operation 520' can improve base interpretation accuracy and reliability, and reduce the computational complexity of prediction, freeing up storage space and saving power and time.

[0274] In some embodiments, compared to training the first neural network with flow pool images from multiple color channels in operation 520', training the second neural network for each corresponding color channel in operation 520' may require less computation, less power consumption, less memory or data storage, reduce training time, and avoid possible training failures.

[0275] Figure 30A An exemplary flow cell image of a first plurality of flow cell images is shown. In this case, the exemplary flow cell image is from a 2D sequencing sample and is obtained from one of four different color channels. The image size is 608 pixels by 608 pixels. Figure 30B for Figure 30A A high-resolution image of the flow cell image is provided, and it is predicted using the second neural network in Method 500 and Operation 520 of this paper. The neural network is pre-trained using a training dataset that includes training flow cell images. The neural network is pre-trained using the training dataset and reference base interpretation results instead of reference strengths. The high-resolution image is 1216 pixels by 1216 pixels in size and provides [images] in the x and y directions. Figure 30A Twice the resolution of the flow cell image. Compared to the first multiple flow cell images (e.g., Figure 30A Compared to high-resolution images (e.g., Figure 30B The detectable polymerase clone density in the image increased by at least 2, 4, or more times. In this case, the high-resolution image predicted by the neural network has less background noise, blur, bright spots, etc. This high-resolution image can then be combined with other high-resolution images from the other three color channels to determine the base interpretation results. In this case, compared with using... Figure 30A Compared to the flow cell image in the middle, using Figure 30B High-resolution images can reduce the error rate in identifying polymerase clones and in making base interpretation results. Predicting high-resolution images instead of directly predicting base interpretation results can advantageously reduce the computational complexity, computational burden, power consumption, and storage space required to perform base interpretation, while maintaining or improving accuracy and reliability.

[0276] In some embodiments, method 500 includes operation 540 (iv) determining polymerase clones from a second plurality of flow cell images using a processor, a first reconfigurable logic device, or an integrated circuit. In some embodiments, determining polymerase clones includes determining the location of a polymerase clone, the center location of a polymerase clone, the size of a polymerase clone, or a combination thereof. In some embodiments, the location or center location of a polymerase clone may be 2D or 3D. In some embodiments, the polymerase clones do not include duplicate polymerase clones. In some embodiments, method 500 includes operation 550: (v) performing a corresponding base readout on each of the determined polymerase clones based on the second plurality of flow cell images using a processor, a first reconfigurable logic device, or an integrated circuit.

[0277] In some embodiments, the second neural network in operation 520' may include one or more layers, such as convolutional layers, during training to generate base interpretation results from the image-processed high-resolution image. In these embodiments, after training, the second neural network in operation 520' may utilize only a subset of the layers of the second neural network being trained, because the neural network only predicts the image-processed high-resolution image and not the base interpretation results. In some embodiments, the second neural network pre-trained using reference base interpretation results may have a first number of layers, while a second number of layers in the pre-trained second neural network is used to predict the high-resolution flow cell image in operation 530. The second number of layers is less than the first number of layers. For example, the pre-trained second neural network may have 5 layers, where the first 4 layers are used to predict the image-processed high-resolution image, and the last layer is used to predict the base interpretation results. In operation 530, only the first 4 layers of the pre-trained second neural network are used. Operation 550 can use the last layer of the pre-trained second neural network.

[0278] In some embodiments, the second neural network in operation 520' utilizes the same number of layers as the second neural network being trained. In such embodiments, the neural network in operation 520' may lack any neural network layers dedicated to generating base interpretation results based on the image-processed high-resolution image. In such embodiments, the neural network in operation 520' may rely on other non-neural network-based algorithms or software to perform base interpretation on the image-processed high-resolution image. For example, the pre-trained second neural network may have five layers, with the first four layers used to predict the image-processed high-resolution image and the last layer used to predict the base interpretation result. In operation 530, only the first four layers of the pre-trained second neural network are used. Operation 550 may use non-neural network-based algorithms or software for base interpretation.

[0279] In some embodiments, the number of convolutional layers in the neural network in operation 520' is less than the number of convolutional layers in the neural network in operation 520. In some embodiments, the number of layers in the second neural network in operation 520' is the same as the number of layers in the first neural network in operation 520.

[0280] In some embodiments, the neural network of this invention has 18, 15, 12, 10, 8, 7, 6, 5, 4, 3, or 2 layers. In some embodiments, the neural network of this invention has 6, 5, 4, 3, or 2 layers. In some embodiments, the neural network has less than 256, 128, 96, 80, 64, or 32 features.

[0281] In some embodiments, method 500 is performed during cycle N, which may be one of the reference cycles used to generate the polymerase clone map. In some embodiments, cycle N may be different from the reference cycle. The polymerase clone map may be generated in the reference cycle as a follow-up operation after the method herein has improved the detectable polymerase clone density in the flow cell image. Polymerase clones from one or more channels within the reference cycle may be included in the polymerase clones in the reference coordinate system, where base interpretation in cycle N has not yet been performed. In some embodiments, cycle N is the current cycle. N can be any non-zero integer. For example, for short read sequencing, N can be any integer from 1 to 150, 1 to 200, or 1 to 1000.

[0282] In some embodiments, the polymerase clone maps disclosed herein may include sub-blocks or individual regions within a block. Each polymerase clone map may include multiple polymerase clones. In some embodiments, the size of the polymerase clone map may be approximately the same as the size of the flow cell image, such that all polymerase clones from different blocks and from multiple channels can be registered to the same polymerase clone map. However, such polymerase clone maps may contain polymerase clones that are not used in at least some of the operations described herein to reduce computational burden without sacrificing accuracy. In some embodiments, more than one polymerase clone map may be generated, and each polymerase clone map corresponds to at least a portion of a sub-block of a flow cell image from a channel. More than one polymerase clone map may be tiled together to cover the entire sample area of ​​the flow cell device.

[0283] In some embodiments, the polymerase clone maps disclosed herein may include polymerase clones within a single cell or tissue or on its membrane. In some embodiments, the polymerase clone maps disclosed herein may not include polymerase clones or signal points located outside cell boundaries. In some embodiments, the polymerase clone maps disclosed herein may not include duplicated polymerase clones, such duplicates occurring at different z-positions, having one or more focused and / or defocused polymerase clones in the flow cell image. Duplicate polymerase clones may be within the same flow cell image or in different flow cell images.

[0284] The polymerase clone map in this paper can be initialized as a dummy image with a black or dark background and no signal from the polymerase clone. For example, the polymerase clone map can be initialized to zero, or a minimum image intensity can be included at all pixels in other ways.

[0285] After determining the coordinates of the polymerase clones through image registration, for example, across flow cell images from different channels, the intensity of the polymerase clones can be added to the polymerase clone map at the locations determined by those coordinates, and its size and shape are determined based on the registration. The polymerase clone map can be a virtual image that combines image intensities from polymerase clones obtained from two, three, four, or even more channels in a reference cycle. Pixels in the template that do not contain polymerase clones remain black or dark, allowing the polymerase clone map to have a cleaner background and free of noise that would appear in the actual flow cell image. In some embodiments, the polymerase clone map includes a list of entries, and each entry corresponds to information used to identify the corresponding polymerase clone. For example, each entry may include the spatial coordinates of the center of the corresponding polymerase clone in the reference coordinate system and the image intensity of the polymerase clone. The entry may also include a unique identifier for the polymerase clone.

[0286] The polymerase clone may be derived from a sub-block of the flow cell image within a reference cycle, and more specifically, from one or more selected regions of that sub-block. The flow cell image may be derived from different channels of one, two, three, four, or more channels of system 100. As a non-limiting example, the reference cycle may be any of the first five or six cycles. In some embodiments, the reference cycle may be any cycle greater than zero. In some embodiments, the reference cycle is the first cycle.

[0287] In some embodiments, operation 540 includes performing an image processing step to adjust the image intensity of the polymerase clone. In some embodiments, the image processing step includes one or more of the following: background subtraction; image sharpening; intensity offset adjustment; color correction; intensity normalization; hysteresis and anti-hysteresis correction; image registration; quality score estimation, etc. In some embodiments, image registration is configured to align images from different cycles and / or different channels, e.g., relative to a template image (i.e., the polymerase clone map) or a reference coordinate system. In some embodiments, image registration herein is configured to register polymerase clones or clusters from different cycles and different channels to a template image or a reference coordinate system.

[0288] In some embodiments, the second plurality of flow pool images may be the output of a neural network. In one or more spatial dimensions, the second resolution may be 2 to 32 times larger than the first resolution. In 2D or 3D, the second resolution may be 4 to 32 times larger than the first resolution.

[0289] In some embodiments, operation 540 is based on a generated polymerase clone map. The polymerase clone map may be 2D or 3D. In some embodiments, the polymerase clone map has a second resolution. In some embodiments, operation 540 includes generating a polymerase clone map and determining a polymerase clone based on the generated polymerase clone map. Details of generating 2D or 3D polymerase clone maps are disclosed in U.S. Patent Applications Nos. 18 / 078,820 and 18 / 078,797, which are incorporated herein by reference in their entirety.

[0290] For example, after registering a second plurality of flow cell images from different channels relative to the polymerase clone map disclosed herein, base interpretation can be performed using the polymerase clone positions in the second plurality of flow cell images from different channels in cycle N. Various existing 2D base interpretation algorithms can be used. The base interpretation results can be saved along with their 3D coordinates. Such 3D coordinates can be used to register base interpretations across different cycles and at different z-levels.

[0291] Method 500 may include the following operation 550(v) performing a corresponding base reading on each of the identified polymerase clones via a processor. The base reading operation 550 may be based on a second plurality of images generated in operation 530. Operation 540 may be further based on the polymerase clone map determined in operation 540. The base reading may be performed using the intensity of polymerase clones from different channels per z-level per cycle.

[0292] In some embodiments, method 500 may include saving the base reading results obtained in operation 550 in a predetermined format (e.g., a FastQ file compatible with subsequent operations) to enable subsequent analyses, such as linker trimming and secondary analysis.

[0293] In some embodiments, the neural network is a convolutional neural network (CNN). In some embodiments, the neural network is a U-Net. In some embodiments, the neural network includes a U-Net having a first predetermined repetition of downsampling and convolution operations and a subsequent second predetermined repetition of upsampling, concatenation, and convolution operations. The first predetermined repetition and the second predetermined repetition may have the same number, for example, 3 or 4 times. In some embodiments, the neural network is a U-Net having a first predetermined number of filters in each downsampling repetition and then a second predetermined number of filters in each upsampling and / or concatenation repetition. For example, in three repetitions, the first predetermined number of filters may be 32, 64, 128, and 256 filters, while in the corresponding three repetitions, the second predetermined number of filters may be 128, 64, 64, and 32 filters, respectively. As another example, in three repetitions, the first predetermined number of filters may be 32, 64, 128, and 256 filters, while in the corresponding three repetitions, the second predetermined number of filters may be 256, 128, 64, and 32 filters, respectively.

[0294] In some embodiments, operation 530 may include: performing a first convolution on a first plurality of flow pool images in one or more dimensions by a processor, thereby generating a first convolution result; repeating a downsampling operation one or more times, including: (a) performing a second convolution on the first convolution result in one or more dimensions by a processor, thereby generating a second convolution result; and (b) downsampling the second convolution result by a downsampling factor by a processor, thereby generating a first downsampled result. In each repetition, the second convolution may include a corresponding number of filters, thereby generating a third convolution result after the repetition.

[0295] In some embodiments, operation 530 may further include: performing a second convolution on the third convolution result in one or more dimensions by a processor, thereby generating a fourth convolution result; repeating an upsampling operation comprising: (c) upsampling the fourth convolution result by an upsampling factor by a processor, thereby generating a first upsampling result; and (d) performing a second convolution on the first upsampling result in one or more dimensions by a processor, thereby generating a fifth convolution result. In each repetition, the second convolution may include a corresponding number of filters, thereby generating a sixth convolution result after the repetition.

[0296] In some embodiments, the first convolution includes a 3D convolution with a kernel. In some embodiments, the kernel may have four dimensions. In some embodiments, the first three spatial dimensions of the kernel are m. m The size of m, its fourth dimension, is determined by the number of filters in the corresponding repetition. In some embodiments, m can be an integer between 2 and 20. For example, the input can be a 512x512 flow pool image, and the z-stack can have 12 slices. The first convolution can include 32 filters, and each filter has a 3x3x3x1 convolution kernel. The output of this convolution block is 512x512x12x32. Then there is a double convolution block, that is, the second convolution has two first convolutions with 32 filters each. The input provided to both blocks is 512x512x12x32, and the output is 512x512x12x32. Each filter uses a convolution kernel of size 3x3x3x3x32. The number of filters can correspond to the features of the input.

[0297] In some embodiments, the second convolution comprises two 3D convolutional layers, for example, as shown in the pseudocode. In other words, the second convolution comprises two repetitions or blocks of the first convolution in 3D, and the number of filters used and applied to the output varies as the convolution process increases the depth of the image. The image depth may increase with the number of features or filters. In some embodiments, the first and second resolutions are 2D or 3D.

[0298] In some embodiments, the first convolution comprises a 2D convolution with a kernel. In some embodiments, the kernel may have three dimensions. In some embodiments, the first two spatial dimensions of the kernel are m x m, and the size of its third dimension is determined by the number of filters in the corresponding repetition. In some embodiments, m may be an integer between 2 and 20. For example, the input may be a flow pool image of size 512x512x1. The first convolution may include 64 filters, and each filter has a 3x3x1 kernel. The output of this convolution block is 512x512x64. Then there is a double convolution block, i.e., the second convolution has two first convolutions with 32 filters each. The input provided to both blocks is 512x512x64, and the output is 512x512x32. Each filter may use a 3x3x32 kernel.

[0299] In some embodiments, the second convolution comprises two convolutional layers, for example, as shown in the pseudocode. In other words, the second convolution comprises two repetitions or blocks of the first convolution, and the number of filters used and varied in the output because the convolution process increases the depth of the image. The image depth can increase with the number of features or filters. In some embodiments, the first and second resolutions are 2D or 3D.

[0300] In some embodiments, the second convolution in operation (a) includes corresponding numbers n, 2, ..., 2 in the first repetition, second repetition, third repetition, and fourth repetition, respectively. n、4 n and 8 n filters. In some embodiments, the second convolution in operation (c) includes the corresponding number of 2 in the last repetition, the penultimate repetition, the penultimate repetition, and the penultimate repetition, respectively. n、2 n、4 n and 8 n filters. In some embodiments, n can be an integer ranging from 8 to 256. For example, operation (a) includes 32, 64, 128, and 256 filters in three repetitions, and operation (c) includes 128, 64, 64, and 32 filters in the corresponding three repetitions.

[0301] In some embodiments, the second convolution in operation (c) includes the corresponding numbers n, 2 in the last repetition, the penultimate repetition, the penultimate repetition, and the penultimate repetition, respectively. n、4 n and 8 n filters. For example, operation (a) includes 32, 64, 128, and 256 filters in four repetitions, and operation (c) includes 256, 128, 64, and 32 filters in the corresponding four repetitions.

[0302] In some embodiments, the second convolution in operation (c) includes the corresponding numbers n, 2, and 2 in the last repetition, the penultimate repetition, and the penultimate repetition, respectively. n、4 n filters. For example, operation (a) includes 32, 64, and 128 filters in three repetitions, and operation (c) includes 128, 64, and 32 filters in the corresponding three repetitions.

[0303] In some embodiments, operation 530 may further include: performing a first convolution on the sixth convolution result in one or more dimensions by a processor, thereby generating a seventh convolution result; and processing to predict a second plurality of flow pooling images based on the seventh convolution result. Each of the second plurality of flow pooling images may correspond to a corresponding flow pooling image in the first plurality of flow pooling images, having a second resolution that is 2, 4, 6, 8, 10, 12, or 16 times larger than the first resolution in one or more spatial dimensions. In some embodiments, the second resolution is at least 4, 6, or 8 times larger than the first resolution in all three dimensions.

[0304] In some embodiments, the first plurality of flow cell images are derived from a single color channel. In some embodiments, the first plurality of flow cell images are derived from one or more color channels. In some embodiments, the first plurality of flow cell images have unbalanced nucleotide diversity across one or more sequencing cycles. In some embodiments, the cell sample contains a spatial density ranging from 10-1. 2 Up to 10 15 Each mm 2 Overloaded multiply molecules. In some embodiments, the cell sample contains a spatial density ranging from 102 3 Up to 10 10 Each mm 2 Overloaded multiply molecules.

[0305] In some embodiments, the first resolution ranges from 0.1 μm to 5 μm. In some embodiments, the first resolution ranges from 0.01 μm to 10 μm. In some embodiments, the second resolution ranges from 0.02 μm to 2 μm. In some embodiments, the second resolution ranges from 0.001 μm to 3 μm. In some embodiments, the downsampling factor is 2, 4, 6, 8, 16, or greater. In some embodiments, the upsampling factor is 2, 4, 6, 8, 16, or greater.

[0306] In some embodiments, one or more of operations (ii) to (v) are performed concurrently with the sequencing run. In some embodiments, one or more operations (ii) to (v) are performed in parallel with the corresponding sequencing run to reduce sequencing analysis time.

[0307] In some embodiments, one or more cycles include the current cycle N. N can range from 1 to 150, 1 to 300, 1 to 500, or 1 to 1000. In some embodiments, one or more of operations (ii) to (v) are performed concurrently with sequencing reactions in cycles following the current cycle N that have not yet been performed or are being performed.

[0308] In some embodiments, the training dataset for training flow cell images comprises a z-stacking of training flow cell images acquired at different z-positions. Each z-stacking may represent a single FOV of a cell sample. In some embodiments, the z-axis is orthogonal to the image plane of the flow cell image.

[0309] In some embodiments, the training dataset for training flow cell images includes flow cell images from multiple sequencing cycles. One or more sequencing cycles may have unbalanced diversity, resulting in images appearing darker or having fewer polymerase clones than images from sequencing cycles with high nucleotide diversity. In other words, the number of polymerase clones in the training flow cell images of a particular cycle can vary from 1% to 99% of the total number of polymerase clones within that cycle's FOV. When the number of polymerase clones in the training flow cell images of a particular cycle accounts for 1% to 5% or 1% to 10% of the total number of polymerase clones in that cycle, its diversity is low or unbalanced. When the number of polymerase clones in the training flow cell images of a particular cycle accounts for more than 10% or 15% of the total number of polymerase clones in that cycle, its diversity is high or unbalanced.

[0310] In some embodiments, the training dataset for training flow cell images includes flow cell images from multiple samples and multiple sequencing cycles, and the training flow cell images include a subset of flow cell images with unbalanced diversity from multiple sequencing cycles and another subset of flow cell images with balanced diversity from multiple sequencing cycles.

[0311] In some embodiments, training flow pool images from one or more cycles may be transformed from other training flow pool images from different cycles to simulate transformations that can occur across cycles within the same color channel.

[0312] In some embodiments, the operation of performing a first convolution on the first plurality of flow pool images in one or more dimensions by a processor includes: performing a first convolution on the first plurality of flow pool images in 3D by a processor, thereby generating a first convolution result. In some embodiments, operation (a) includes performing a second convolution on the first convolution result in 3D by a processor, thereby generating a second convolution result.

[0313] In some embodiments, the operation of performing a first convolution on the first plurality of flow pool images in one or more dimensions by a processor includes: performing a first convolution on the first plurality of flow pool images in 2D by a processor, thereby generating a first convolution result. In some embodiments, operation (a) includes performing a second convolution on the first convolution result in 2D by a processor, thereby generating a second convolution result.

[0314] In some embodiments, repeating the operations including (c) and (d) once or more includes: repeating the operations including (c), (d), and (e) once or more, wherein (e) follows operation (c) and precedes operation (e), and wherein (e) includes: concatenating a first upsampling result in the current upsampling repetition with a first downsampling result in the previous downsampling repetition by a processor, wherein the first upsampling result and the first downsampling result in the previous downsampling repetition have the same size. In some embodiments, operation (e) occurs in each repetition. In other words, repeating the operations including steps (c) and (d) once or more includes: repeatedly performing the operations including steps (c), (d), and (e) in each of the one or more repetitions.

[0315] The convolution kernel can be any size smaller than the size of the flow pool image to which convolution is performed. For example, using an opening operation, the convolution kernel can be 2x2x2, 3x3x3, 4x4x4, 5x5x5, or 6x6x6 in the first three spatial dimensions. In some embodiments, the convolution kernel size can be customized to remove at least a portion of noise and unwanted signals larger than the kernel size. In some embodiments, the convolution kernel can be ring-shaped. Convolution kernels can take on a variety of other shapes.

[0316] In some embodiments, the focal point of the optical system includes a range extending along the z-axis, for example, 0.1 μm, 0.2 μm, 0.3 μm, 0.5 μm, 0.6 μm, 0.8 μm, 1 μm, 2 μm, 3 μm, 4 μm, 5 μm, etc. Polymerase clones or clusters within this focal range may appear in focus or near focus in the flow cell image. Flow cell images at a specific z-level may also include signals from polymerase clones or clusters located outside the image's focal range but at different z-levels. Such polymerase clones or clusters are out of focus. Figure 3A As shown, larger and blurry signal points represent out-of-focus polymerase clones or clusters. Figure 3A Some of the defocused polymerase clones or clusters are circled in the middle.

[0317] Each flow cell image at a specific z-level may also include noise caused by the optical system and / or unwanted signals from the sample. Unwanted signals can be signals from sample components such as cell membranes, cytoplasm, and mitochondria. Such background objects can be any object that is relatively large in size compared to a polymerase clone or cluster. Figure 3A As shown, the flow cell image contains blurred cell outlines (indicated by arrows), and most signal points are contained within these blurred outlines. In some embodiments, background objects may include any object that is within the 3D sample but is not a polymerase clone or cluster.

[0318] In some embodiments, method 500 includes the operation of registering a second plurality of flow cell images. In some embodiments, images are registered across channels and / or across different cycles. In some embodiments, images are registered prior to performing any base interpretation in operation 550. In some embodiments, images are registered across channels and different cycles prior to generating or obtaining a polymerase clone map. In some embodiments, images are registered across channels and different cycles prior to performing one or more major analytical steps herein. In some embodiments, images may be registered after performing one or more preprocessing operations disclosed herein. Various image registration techniques can be used to register images. Various image registration techniques can be used to register images. 2D or 3D registration techniques can be used to register images.

[0319] In some embodiments, the operation of registering the flow cell image is performed relative to a reference coordinate system. In some embodiments, the operation of registering the flow cell image is performed relative to one or more template images. The operation of registering the image may include generating one or more template images in the reference coordinate system. In some embodiments, the operation of registering the image may include registering a polymerase clone to a template polymerase clone in one or more template images. The operation of registering the image may include determining a plurality of transformations based on one or more template images. Each of the plurality of transformations may correspond to a corresponding sub-block of the flow cell image, the processed image, or the filtered image, and is configured to register that sub-block to one or more template images. Each transformation can be used to register the corresponding sub-block or block to one or more template images. The plurality of transformations may include one or more affine transformations.

[0320] In some embodiments, the operation of registering images may include performing image registration on polymerase clones based on a reference marker. The reference marker may be located on the flow cell. Alternatively, the reference marker may be external to the flow cell.

[0321] In some embodiments, image registration, as an image processing step herein, is configured to align images from different cycles and / or different channels, for example, relative to a template image or a reference coordinate system. In some embodiments, image registration herein is configured to register polymerase clones or clusters from different cycles and / or different channels (e.g., in a filtered image) to a template image or a reference coordinate system.

[0322] For example, after registering filtered images from different channels with the corresponding template images disclosed herein, base reading can be performed using filtered images from different channels in cycle N.

[0323] Operation 540 may include an operation to extract polymerase clone intensity based on a polymerase clone map. For each polymerase clone in the polymerase clone map, location information of that polymerase clone, such as the 2D coordinates and z-level of the polymerase clone, can be obtained from the polymerase clone map. Using the 2D coordinates and z-level, the corresponding flow cell image and its pixels can be determined. After one or more image processing steps, the image intensity of such pixels can be extracted from the corresponding processed image as the intensity of such pixels for performing base determination.

[0324] In some embodiments, the registration of flow cell images can be based on background objects in the flow cell image. The flow cell image can be aligned with the cell image using one or more transformations based on the background objects. The cell staining image described herein is a staining image of a sample fixed to a support, with possible transformations (e.g., translations) relative to the sample in the flow cell image. This transformation can be represented as a single transformation of the entire image, or it can be decomposed into multiple transformations, each representing a portion of the entire image. After finding the transformations of the background objects between the flow cell image and the cell staining image, polymerase clones or clusters can be registered to the cell staining image.

[0325] In some embodiments, method 500 may include the operation of registering base interpretations in 550 to a cell staining image. In some embodiments, such registration may be based on a reference marker. Such a reference marker may also be included in the cell staining image. Aligning the reference marker may generate transformations between flow cell images or between a flow cell image and a cell staining image. The transformations may be used to register or align polymerase clones or clusters between sequencing images and cell images.

[0326] For example, the simulated z-stacking is 2048x2048x3, and each cell can contain 200 to 2000 polymerase clones. The spatial resolution can be approximately 0.1 μm. Prediction is performed independently for each 512x512 region of the simulated z-stacking. The predicted high-resolution z-stacking is 8192x8192x12. Figures 2A to 2C The simulated flow cell image is shown, along with two different predicted flow cell images at different z-positions with 4x resolution. Figure 3A and Figure 3D The image shows two actual flow cell images at different z positions in a 512x512x3 z stack. Figures 3B to 3C The predicted high-resolution flow cell image (2048x2048) is in the corresponding Figure 3A At two different z-positions in the low-resolution image.

[0327] Exemplary pseudocode for predicting high-resolution flow cell images

[0328] An exemplary neural network is illustrated in pseudocode below. The neural network can be used to predict polymerase cloning locations using a z-stack comprising flow cell images from multiple z-levels forming a 3D volume. In some embodiments, m ranges from 2 to 10, the filters range from 8 to 1024, k_size can be in four dimensions, and the fourth dimension of k_size can match the number of filters in the corresponding repetition. As disclosed herein, the input flow cell image can have various sizes in 3D, for example, 1024 x 1024 x 4.

[0329]

[0330] Exemplary pseudocode for predicting high-resolution flow cell images

[0331] An exemplary neural network is illustrated in pseudocode below. This neural network can be used to predict polymerase clone locations based on 2D flow cell images at different z-levels. In some embodiments, m ranges from 2 to 10, the filters range from 8 to 1024, k_size can be in three dimensions, and the third dimension of k_size can match the number of filters in the corresponding repetition. As disclosed herein, the input flow cell image can have various sizes in 2D, for example, 1024 by 1024, and there can be 3, 4, 5, or other numbers of z-levels.

[0332]

[0333] Using neural networks to predict base interpretation results

[0334] In some embodiments, the methods and systems of this invention can be used to predict base interpretation results for some or all polymerase clones in a flow cell image. The systems and methods of this invention advantageously utilize a neural network pre-trained to predict base interpretation results for polymerase clones in a flow cell image. Without additional training, the same neural network can also be advantageously used to generate polymerase clone maps or template images, allowing the location of the predicted base interpretation results to be determined. The embodiments of this invention use convolutional neural networks as examples; however, it should be understood that various other neural networks or machine learning models can also be used to implement the prediction of base interpretation results using the systems and methods of this invention.

[0335] In some embodiments, a method for predicting base interpretation results may include one or more of the operations described herein. When multiple operations are involved, such operations may or may not be performed in the order described herein.

[0336] Figure 28A flowchart of a computer-implemented method 2800 is shown, which is used to predict the base interpretation results of a flow cell image of a biological sample (e.g., a cell sample), thereby achieving efficient and accurate primary analysis. Method 2800 may include some or all of the operations disclosed herein. These operations may be performed in, but are not limited to, the order described herein.

[0337] Method 2800 may be executed by one or more processors disclosed herein. In some embodiments, a processor may include one or more of the following: a processing unit (e.g., a CPU), a reconfigurable logic device, a non-reconfigurable integrated circuit, or a combination thereof. For example, a processing unit may include a central processing unit (CPU). A reconfigurable logic device may include one or more FPGA devices. An integrated circuit may include a chip, such as an AI chip or an ASIC chip. In some embodiments, the processor may include a computing system 400.

[0338] In some embodiments, some or all of the operations in method 2800 may be performed by a reconfigurable logic device (e.g., an FPGA) and / or an integrated circuit (e.g., an AI chip). In embodiments, when some operations are performed by a reconfigurable logic device and / or an integrated circuit (e.g., an FPGA), data generated by the reconfigurable logic device and / or integrated circuit (e.g., an FPGA) after performing one or more operations may be transmitted to various hardware elements of system 100 (e.g., a CPU or GPU), allowing subsequent operations in methods 500, 600, 700, 2800, and 2900 to be performed using the transmitted data via such hardware. Similarly, data may also be transmitted in reverse from various hardware (e.g., a CPU) to a reconfigurable logic device or integrated circuit for processing. In some embodiments, all operations in the methods herein may be performed by a CPU. Alternatively, operations performed by a CPU may be performed by other processors or GPUs, such as dedicated processors. In some embodiments, all operations in the methods herein may be performed by a reconfigurable logic device and / or an integrated circuit (e.g., an FPGA and / or an AI chip).

[0339] In some embodiments, sensor data acquired by imager 116 can be directly transmitted to a reconfigurable logic device and / or integrated circuit (e.g., via a DMA connection). In some embodiments, sensor data acquired by imager 116 can be directly transmitted to a reconfigurable logic device and / or integrated circuit without first being routed to a CPU, GPU, or any other processing unit before reaching the reconfigurable logic device and / or integrated circuit.

[0340] In some embodiments, the power required to make predictions or inferences using the method 2800 described herein with reconfigurable logic devices (e.g., FPGAs) and / or other integrated circuits (e.g., AI chips) may be at least 2, 8, 10, 15, 20, 40, 50, or 100 times lower than the power required to make predictions or inferences using other computing hardware (including but not limited to CPUs or GPUs) with the same neural networks and the same training images.

[0341] In some embodiments, the sequencing system of this invention further includes a power supply configured to provide the same or different power levels to the reconfigurable logic devices and integrated circuits. In some embodiments, when performing methods 500, 600, 700, 2800 and / or 2900, the maximum power output provided by the power supply to the sequencing system is less than 2000 watts, 1000 watts, 900 watts, 800 watts, 700 watts, 650 watts, 600 watts, 550 watts, 500 watts, 400 watts, 300 watts, 200 watts, or 100 watts.

[0342] Method 2800 may include the following operations 2810: (i) generating a first plurality of flow cell images of a sample immobilized on a support by performing one or more cycles of sequencing reactions using the sequencing system 110.

[0343] The sample can be a conventional 2D sequencing sample containing a bioanalyte. The sample can be a cell or tissue sample. The sample may contain polymolecules. The sample may contain polymolecules from one or more different sample sources. The sample may have a certain thickness along the z-axis, enabling the acquisition of first plurality of flow cell images at z-stacks at different z-positions with a first resolution to cover the cell sample in 3D.

[0344] The sample can be in situ. The sample can be a 3D sample. The sample can be a volumetric sample, which may contain different biological information at the same xy location but different z-levels. The sample may include multiple cells, tissues, or combinations thereof. A 3D sample can be any biological sample with a thickness along the z-axis greater than a predetermined threshold. For example, the thickness may be greater than 1 μm, 2 μm, 3 μm, 4 μm, 5 μm, 10 μm, 20 μm, or greater. The z-axis (e.g., the z-axis line) is orthogonal to the image plane defined by the x-axis and y-axis. In some embodiments, the sample can be a conventional 2D sequencing sample.

[0345] Flow cell images can be acquired from 1, 2, 3, 4 or more channels using the optical system of the imager 116 disclosed herein. Each flow cell image may include at least a portion of one or more blocks (e.g., imaging regions), and each block may be subdivided into multiple sub-blocks. Each block or sub-block may include multiple polymerase clones or clusters. Each sub-block may include multiple regions, wherein each region includes several polymerase clones. Flow cell images as disclosed herein can be acquired from, for example,... Figure 1 The flow cell 112 shown is or Figure 27 The image obtained by 2712 shown is illustrated. In some embodiments, the flow pool image is obtained from a single color channel, and subsequent prediction is performed using a pre-trained neural network corresponding to that single channel. In some embodiments, the flow pool image is obtained from 2, 3, 4, or more color channels, and subsequent prediction is performed using a pre-trained neural network corresponding to those multiple color channels.

[0346] In some embodiments, the flow cell image herein may be an image of one or more blocks, one or more sub-blocks, one or more segmented regions within a block or sub-block, or a combination thereof. Each flow cell image may include a field of view (FOV). The FOV may be perpendicular to the z-axis. The FOV may lie in the xy-plane. The FOVs of different flow cell images at different z-levels may be the same in the xy-plane. The FOVs of different flow cell images at different z-levels may have at least an overlap in the xy-plane. The image resolutions of different flow cell images at different z-levels may be substantially the same or identical. In some embodiments, the image resolutions of different flow cell images at different z-levels are different. Figure 3A and Figure 3D Two exemplary flow cell images acquired at two different z-levels along the z-axis of the same 3D sample within the same sequencing cycle are shown. The field of view (FOV) can be 3D and can have various sizes to cover the volume of sample to be imaged. The FOV can range from 10 μm to 5 mm along the x, y, and / or z directions. The FOV can range from about 0.1 μm to about 2 mm along the x, y, and / or z directions. The FOV can range from 0.5 μm to 1 mm along the x, y, and / or z directions. For example, for some cell samples, the FOV can be about 0.5 mm x 0.5 mm x 20 μm along the x, y, and z directions, respectively.

[0347] The flow cell images described herein can have various sizes, and the number of pixels along the x-axis, y-axis, and / or z-axis can be any integer greater than 64 or 128. The number of pixels along the x-axis, y-axis, and / or z-axis can range from 2 to 65,536. A single flow cell image can be divided into different numbers of regions, such as 4, 8, 16, or even more regions, each region's size can be 256 x 256 x 1, 512 x 512 x 3, or other sizes. In some embodiments, the number of pixels along the x-direction, y-direction, and / or z-direction can be adjusted to maintain a specific spatial resolution within a given FOV. For example, for a spatial resolution of 0.2 μm, the number of pixels could be 4000 to cover an FOV of 0.8 mm.

[0348] Each flow cell image at a specific z-level can contain intensities generated by polymerase clones or clusters at the corresponding z-level. For example... Figure 3A and Figure 3D As shown, the signals from polymerase clones or clusters are small bright spots within the image. Each bright spot can be of various sizes, less than a few pixels, for example, less than one pixel, about one pixel, about two pixels, three pixels, four pixels, five pixels, or larger. In some embodiments, each signal point of a polymerase clone or cluster can be any number of pixels ranging from 0.01 pixels to about 100 pixels. In some embodiments, each signal point of a polymerase clone or cluster can be any number of pixels ranging from 0.1 pixels to about 16 pixels.

[0349] Each flow cell image may also include intensities generated by cells and their structural elements. Such structural elements can be background objects or components, for example, in... Figure 3A In addition, each flow cell image may also include noise and / or artifacts that are not derived from polymerase clones or cell structures.

[0350] In some embodiments, when the depth of field of the optical system includes a range extending along the z-axis (e.g., 0.1 μm, 0.2 μm, 0.3 μm, 0.5 μm, 0.6 μm, 0.8 μm, 1 μm, 2 μm, 3 μm, 4 μm, 5 μm, etc.), polymerase clones or clusters within this depth of field may appear in focus or nearly in focus in the flow cell image. Flow cell images at specific z-levels may also include signals from polymerase clones or clusters that are not within the image's focus range. Such polymerase clones or clusters are out of focus. Figure 3A As shown, larger and blurry signal points represent out-of-focus polymerase clones or clusters. Figure 3A Some of the defocused polymerase clones or clusters are circled in the middle.

[0351] Each flow cell image at a specific z-level may also include noise caused by the optical system and / or unwanted signals from the sample. Unwanted signals can be signals from sample components such as cell membranes, cytoplasm, and mitochondria. Such background objects can be any object that is relatively large in size compared to a polymerase clone or cluster. Figure 3A As shown, the flow cell image contains blurred cell outlines (indicated by arrows), and most signal points are contained within these blurred outlines. In some embodiments, background objects may include any object that is within the 3D sample but is not a polymerase clone or cluster.

[0352] In some embodiments, the base interpretation results from the polymerase clones include four different bases, and for each of the four bases, the percentage of the polymerase clone may be greater than about 10%, making the data relatively diverse. In some other embodiments, the base interpretation results from the plurality of polymerase clones include four or fewer different bases, and for one or more bases, the percentage of the polymerase clone may be less than about 10%, and such data can be considered data with unbalanced diversity. In some embodiments, the base interpretation results from the plurality of polymerase clones include four or fewer different bases, and for some of the bases, the percentage of the polymerase clone may be less than about 5%, about 2%, or even about 1%, and such data can be considered data with unbalanced diversity. As an example, in the plurality of polymerase clones, the base interpretation results for bases A, T / U, C, and G may be about 1%, about 2%, about 1%, and about 95%, respectively. As another example, in these multiple polymerase clones, the base interpretation results for A, T / U, C, and G can be approximately 10%, approximately 10%, approximately 10%, and approximately 70%, respectively. Besides base bias affecting diversity, complexity is also a factor; when the complexity is below a certain number (e.g., 8 or 16), the signal may exhibit unbalanced diversity. Method 2800 is configured to predict (e.g., at a first resolution) the base interpretation results of a flow cell image, even if the polymerase clones in the flow cell image exhibit unbalanced nucleotide diversity across one or more sequencing cycles, and to spatially align the base interpretation results with polymerase clones in a flow cell image at a second resolution. The second resolution may be higher than the first resolution.

[0353] In some embodiments, method 2800 includes the following operation 2802: (ia) generating a second plurality of flow pool images including a second resolution via a processor or a first reconfigurable logic device. In some embodiments, each of the second plurality of flow pool images corresponds to a corresponding flow pool image in the first plurality of flow pool images. The second plurality of flow pool images may be generated using various upsampling algorithms (including, but not limited to, interpolation). The second resolution may be greater than the first resolution in one or more spatial dimensions. The second resolution may be at least twice the size of the first resolution in one or more spatial dimensions. The second resolution may be two to 32 times the size of the first resolution in one or more spatial dimensions. The second resolution may be four to 64 times the size of the first resolution in one or more spatial dimensions (e.g., along the x-direction, y-direction, and / or z-direction). The second resolution may be at least two to 32 times the size of the first resolution in one or more spatial dimensions. The second resolution may be at least four to 64 times the size of the first resolution in one or more spatial dimensions.

[0354] In some embodiments, method 2800 includes the following operation 2804: (ii) providing a second plurality of flow cell images as input to a neural network (e.g., a convolutional neural network (CNN)) via a processor, wherein the neural network is pre-trained using a training dataset of training flow cell images using training methods disclosed herein (e.g., 600, 700, 2900 herein). In some embodiments, the neural network is pre-trained such that the parameter values ​​(e.g., weights) of the neural network have been optimized based on training. The neural network may be retrained if necessary, for example, for predicting flow cell images from different cell samples.

[0355] In some embodiments, method 2800 may include image processing steps performed on a first plurality of flow pool images or a second plurality of flow pool images, which may optionally be performed before any input is provided to the neural network. The processing steps may include: intensity normalization, background subtraction, background removal, artifact reduction, artifact removal, signal-to-noise ratio adjustment, contrast-to-noise ratio adjustment, color correction, intensity offset adjustment, image registration, hysteresis and advance, filtering, segmentation, noise reduction, deconvolution (e.g., for distinguishing adjacent or at least partially overlapping signal points) or combinations thereof.

[0356] In some embodiments, method 2800 includes the following operation 2804': providing a first plurality of flow cell images or a second plurality of flow cell images to a polymerase clone map generation algorithm or a base interpretation algorithm via a processor, a first reconfigurable logic device, or an integrated circuit. In some embodiments, the polymerase clone map generation algorithm and the base interpretation algorithm do not include a trained neural network or an artificial intelligence-based algorithm. Exemplary polymerase clone map generation algorithms for generating 2D or 3D polymerase clone maps and base interpretation algorithms for generating base interpretation results are disclosed in U.S. Patent Applications Nos. 18 / 078,797 and 18 / 078,820 and U.S. Patent No. 10,266,888, which are incorporated herein by reference in their entirety.

[0357] In some embodiments, method 2800 includes the following operation 2806: (iia) determining a polymerase clone map based on a second plurality of flow cell images using a first reconfigurable device or integrated circuit. In some embodiments, operation 2806 may be based on operation 2804, and in some other embodiments, it may be based on operation 2804'.

[0358] In some embodiments, the polymerase clone map is 3D. In some embodiments, the 3D polymerase clone map comprises multiple 2D polymerase clone maps at different z-levels. In some embodiments, the 3D polymerase clone map has a second resolution. In some embodiments, a polymerase clone map generation algorithm is used to generate the polymerase clone map. In some embodiments, the polymerase clone map generation algorithm lacks any neural network or artificial intelligence-based algorithms. In some embodiments, the polymerase clone map generation algorithm lacks any pre-trained neural network that can predict base interpretation results in operation 2812 without additional training for predicting polymerase clone maps. In some embodiments, the polymerase clone map generation algorithm utilizes conventional algorithms lacking artificial intelligence.

[0359] In some embodiments, the neural network in operations 2804 to 2806 is the same pre-trained neural network used in operation 2812. In some embodiments, this same pre-trained neural network may include the same parameters, layers, and neural network structure. In some embodiments, the same pre-trained neural network may include the same number of parameters, the same number of layers, and neural network structure. In some embodiments, method 2800 may further include the operation of training the neural network prior to operations 2804 and 2806. In some embodiments, the neural network is trained prior to operations 2804 and 2806, for example, using methods 700 or 2900 disclosed herein. In some embodiments, the pre-trained neural network can be used to predict polymerase clone location, polymerase clone shape and / or size, polymerase clone center location, or equivalently predict polymerase clone map. In some embodiments, operation 2800 may further include one or more operations in method 500, such as operations 530 and 540, and / or 550, for predicting the location of the polymerase clone, and thus predicting the polymerase clone map.

[0360] In some embodiments, the same training data, including images of the same flow cell from the samples, can be used to train the same neural network used in operations 2804, 2806, and 2812. The same training data may also include the same "benchmark" or reference during training. As a result, after training, the same neural network may include the same parameter values, the same number of layers, and the same neural network structure.

[0361] In some embodiments, the same neural network used in operations 2804, 2806, and 2812 may be trained using at least different portions of the same training data. As a result, this same neural network may include the same parameters, the same layers, and the same neural network structure with the same or different values ​​for such parameters. Training of this same neural network can be performed before operation 2804, and it is not necessary to retrain the neural network after operation 2806 and before operation 2812. The pre-trained neural network can then be used in operations 2804 through 2806 and operation 2812 without retraining, allowing for rapid and efficient prediction of base interpretation results using method 2800.

[0362] In some embodiments, a pre-trained neural network can be used in operations 2804 to 2806 to update an existing polymerase clone map. The existing polymerase clone map may be generated in an earlier cycle of a sequencing run. The polymerase clone map predicted using the pre-trained neural network can be used to update the existing polymerase clone map in subsequent cycles of a sequencing run. For example, an initial polymerase clone map may be generated using a non-neural network algorithm in the first cycle or the first few cycles of a sequencing run (e.g., cycles 1 to 4). Data from the first cycle or several cycles (e.g., cycles 1 to 4 or cycles 1 to 5) can be used to train the neural network. The pre-trained neural network can then be used to predict a second polymerase clone map, which can be used to update the initial polymerase clone map. The second polymerase clone map can advantageously reselect a more accurate and reliable polymerase clone position for making predictions about base interpretation results, strengths, or classifications, for example, in operation 2812. Such predictions can be repeated by training the neural network with different cycles already completed in the sequencing run to improve the reselection of polymerase clone positions. For example, after training using data from cycles 1 to 4, the trained neural network can be retrained using data from cycles 1 to 6 or 1 to 7, and another prediction can be made on the polymerase clone map after training.

[0363] In some embodiments, different reference information can be used as “benchmark truth” during training to train the same neural network used in operations 2806 and 2816. In some embodiments, the training of the neural network for predicting polymerase cloning positions can use reference strength as “benchmark truth”, while the training of the neural network for predicting base interpretation results can use reference base interpretation results as “benchmark truth”.

[0364] In some embodiments, the same reference information can be used as the “benchmark truth” during training to train the same neural network used in operations 2804 to 2806 and 2816. In some embodiments, the training of the neural network for predicting polymerase cloning positions can use a reference strength as the “benchmark truth”, and the training of the neural network for predicting base interpretation results can use a reference base interpretation result that can be determined based on such a reference strength.

[0365] In some embodiments, the same neural network can be trained using various training methods (e.g., method 2900 disclosed herein) to predict base interpretation results. A reference base interpretation result can be used to train the neural network. (e.g., using method 2900) The reference base interpretation result used in training may include its spatial information. (e.g., using method 2900) The reference base interpretation result used in training may have a first resolution, a second resolution, or a third resolution. In some embodiments, the third resolution may be higher than the first and second resolutions.

[0366] In some embodiments, the same neural network may be trained, for example, using method 2900, to predict base interpretation results. After training, the neural network can be used to predict base interpretation results for a second plurality of flow cell images. The predictions of the base interpretation results can then be processed to determine the location of polymerase clones, thereby generating a polymerase clone map. For example, the polymerase clone map may be determined as the location where a base interpretation result is predicted to have a probability of satisfying a predetermined threshold. As another example, the polymerase clone location (and therefore the polymerase clone map) may be determined as the location where one or more quality metrics satisfy a predetermined threshold. Such quality metrics may include, but are not limited to, the maximum intensity, intermediate intensity, or average intensity of the polymerase clone in different color channels, the Q score of the base interpretation result, the sharpness of the base interpretation result, and the purity of the base interpretation result.

[0367] As disclosed above, the second plurality of flow cell images can be used to generate a polymerase clone map at a second resolution using operations 2804 to 2806 or operations 2804' to 2806. Alternatively, in some embodiments, the method may include an operation to generate a polymerase clone map at a first resolution based on the first plurality of flow cell images, and an operation to upsample after operation 2810 to generate a polymerase clone map at a second resolution. In such embodiments, the first plurality of flow cell images may be provided instead of the second plurality of flow cell images in operation 2804 or operation 2804', and then operation 2806 may be replaced with an operation to determine the polymerase clone map based on the first plurality of flow cell images.

[0368] In some embodiments, method 2800 is performed during cycle N, which may be one of the reference cycles used to generate the polymerase clone map. In some embodiments, cycle N may be different from the reference cycle. The polymerase clone map may be generated in the reference cycle as a follow-up operation after the method herein has improved the detectable polymerase clone density in the flow cell image. Polymerase clones from one or more channels within the reference cycle may be included in the polymerase clones in the reference coordinate system, where base interpretation of cycle N has not yet been performed. In some embodiments, cycle N is the current cycle. N can be any non-zero integer. For example, for short read sequencing, N can be any integer from 1 to 150. In some embodiments, N can be any integer from 1 to 20, 1 to 200, 1 to 300, 1 to 500, or 1 to 1000.

[0369] In some embodiments, the polymerase clone maps disclosed herein may include a single region or sub-block within a sub-block. Each polymerase clone map may include multiple polymerase clones. In some embodiments, the size of the polymerase clone map may be approximately the same as the size of the flow cell image, such that all polymerase clones from different blocks and from multiple channels can be registered to the same polymerase clone map. However, such polymerase clone maps may contain polymerase clones that are not used in at least some of the operations described herein to reduce computational burden without sacrificing accuracy. In some embodiments, more than one polymerase clone map may be generated, and each polymerase clone map corresponds to at least a portion of a sub-block of a flow cell image from a channel. More than one polymerase clone map may be tiled together to cover the entire sample area of ​​the flow cell device.

[0370] In some embodiments, the polymerase clone maps disclosed herein may include polymerase clones within a single cell or tissue or on its membrane. In some embodiments, the polymerase clone maps disclosed herein may not include polymerase clones or signal points located outside cell boundaries. In some embodiments, the polymerase clone maps disclosed herein may not include duplicated polymerase clones, such duplicates occurring at different z-positions, having one or more focused and / or defocused polymerase clones in the flow cell image. Duplicate polymerase clones may be within the same flow cell image or in different flow cell images.

[0371] The polymerase clone map presented in this paper can be initialized as a dummy image with a black or dark background and no signal from the polymerase clone. For example, the polymerase clone map can be initialized to zero, or a minimum image intensity can be included at all pixels in other ways.

[0372] After determining the coordinates (e.g., 3D coordinates) of the polymerase clone through image registration, for example, across flow cell images from different channels, the intensity of the polymerase clone can be added to the polymerase clone map at the location determined by those coordinates, and its size and shape are determined based on the registration. The polymerase clone map can be a virtual image that combines image intensities from polymerase clones obtained from two, three, four, or even more channels in a reference cycle. Pixels in the template that do not contain polymerase clones remain black or dark, allowing the polymerase clone map to have a cleaner background and free of noise that would appear in the actual flow cell image. In some embodiments, the polymerase clone map includes a list of entries, and each entry corresponds to information used to identify the corresponding polymerase clone. For example, each entry may include the spatial coordinates of the center of the corresponding polymerase clone in the reference coordinate system and the image intensity of the polymerase clone. The entry may also include a unique identifier for the polymerase clone.

[0373] The polymerase clone may be derived from a sub-block of the flow cell image within a reference cycle, and more specifically, from one or more selected regions of that sub-block. The flow cell image may be derived from different channels of one, two, three, four, or more channels of system 100. As a non-limiting example, the reference cycle may be any of the first five or six cycles. In some embodiments, the reference cycle may be any cycle greater than zero. In some embodiments, the reference cycle is the first cycle.

[0374] In some embodiments, the processing steps described herein include performing the image processing steps described herein to adjust the image intensity of polymerase clones. In some embodiments, the image processing steps include one or more of the following: background subtraction; image sharpening; intensity offset adjustment; color correction; intensity normalization; hysteresis and anti-hysteresis correction; image registration; quality score estimation, etc. In some embodiments, image registration is configured to align images from different cycles and / or different channels, e.g., relative to a template image (i.e., a polymerase clone map) or a reference coordinate system. In some embodiments, the image registration described herein is configured to register polymerase clones or clusters from different cycles and different channels to a template image or a reference coordinate system.

[0375] Method 2800 may include the following operation 2812: (iii) using a neural network via a first reconfigurable device or integrated circuit to predict one or more base interpretation results corresponding to one or more polymerase clones in a second plurality of flow cell images; or using a neural network via a first reconfigurable device or integrated circuit to predict one or more classifications corresponding to one or more pixels in a second plurality of flow cell images. The base interpretation operation 2812 may be based on the second plurality of flow cell images. Operation 2812 may further be based on the polymerase clone map determined in operation 2804 or 2804'.

[0376] In some embodiments, the second plurality of flow cell images may be derived from one or more color channels, one or more z-levels, and / or one or more cycles. The prediction of base interpretation results in operation 2812 can be performed using the strength of polymerase clones. In some embodiments, the second plurality of flow cell images may be derived from a single color channel, a single z-level, and / or a single cycle. In some embodiments, the prediction of base interpretation results can be performed using the strength of polymerase clones from a single color channel and one or more cycles. For example, flow cell images acquired from each of the plurality of color channels in multiple cycles can use different pre-trained neural networks to predict the polymerase clone strength of the corresponding channel. The prediction in operation 2812 of base interpretation can then be performed using the strength of polymerase clones from different color channels. As another example, flow cell images from a single z-level may require different pre-trained neural networks to predict base interpretation results from different z-levels using operation 2812. In some embodiments, the prediction of base interpretation results in operation 2812 can be performed using the strength of polymerase clones from different color channels, multiple z-levels, and multiple cycles. In some embodiments, the prediction of base interpretation in operation 2812 can be performed using the intensity of polymerase clones from different color channels, a single z-level, and multiple cycles. In some embodiments, the prediction of base interpretation in operation 2812 can be performed using the intensity of polymerase clones from a single color channel, one or more z-levels, and one or more cycles. In some embodiments, the prediction of base interpretation in operation 2812 can be performed using the intensity of polymerase clones from one or more color channels, one or more z-levels, and one or more cycles.

[0377] In some embodiments, operation 2812 (iii) may include generating an output that includes base determination results, such as A, T, C, G, and / or U for one or more pixels in a second plurality of flow cell images. The one or more pixels may be determined using a polymerase clone map or a list of polymerase clone locations disclosed herein, such that each of the one or more pixels is included in at least one polymerase clone in the polymerase clone map.

[0378] In some embodiments, operation 2812 of (iii) may include generating output including base interpretation results, such as A, T, C, G, and / or U for one or more pixels in the second plurality of flow cell images. In some embodiments, operation 2812 of (iii) may include generating output including classification, such as A, T, C, G, U, and / or background for one or more pixels in the second plurality of flow cell images. In some embodiments, the one or more pixels may include pixels not included in the polymerase clone map or location list disclosed herein. For example, the one or more pixels may include all pixels within the field of view (FOV) of the second plurality of flow cell images. In some embodiments, the one or more pixels include at least one pixel not included in any polymerase clone in the polymerase clone map. In some embodiments, the one or more pixels include at least one pixel included in the background of the polymerase clone, the background including noise signals. In some embodiments, the one or more pixels include at least one pixel not included in any polymerase clone in the polymerase clone map, and at least one pixel included in at least one polymerase clone in the polymerase clone map. In some embodiments, the one or more pixels include at least one pixel not within or on the cell membrane.

[0379] Figures 3E to 3F A comparison of the accuracy of identifying transcripts (corresponding to polymerase clones) using the neural network method (“Novel Algorithm”) proposed herein (e.g., 2800) and a conventional non-neural network-based algorithm (“POR-YOLO”) is shown. In this case, simulated flow cell images of in situ samples with multiple cells are used. Each region may include 0 to 4000 targets. Such targets may be transcripts. The neural network described herein (e.g., in method 2800) and the classical non-neural network-based algorithm are used to predict / detect transcripts in such cells. The predicted / determined results are then compared to a baseline ground truth (or equivalently, referring to a polymerase clone map) to assess accuracy. The number of correct targets per region is higher using the neural network and method disclosed herein compared to the non-neural network-based algorithm. When the target density per region is greater than 2000 per region, the number of targets detected per region using the neural network and method disclosed herein is significantly higher (2-fold or 3-fold higher) than the number detected by the non-neural network-based algorithm. Figure 3F The false negative rates per cell for both the neural network (“new algorithm”) and the non-neural network-based algorithm (“POR-YOLO”) are shown. When the target density per region is greater than 1000 per region, the false negative rate per region using the neural network and method disclosed herein is significantly lower (10 times or more) than that detected by the non-neural network-based algorithm.

[0380] Figure 3G A comparison of the accuracy of transcript identification using methods (e.g., Method 2800) and conventional non-neural network-based algorithms is shown. In this case, simulated flow cell images of in situ samples with multiple cells are used. Each cell may contain 0 to 6000 transcripts. The neural network described in this paper (e.g., in Method 2800) and classic non-neural network-based algorithms are used to predict / detect transcripts in such cells. The predicted / determined results are then compared to ground truth values ​​(or equivalently, referencing polymerase clone maps) to assess accuracy. R 2 The values ​​show the correlation between the predicted / determined result and the reference value. Compared to classical non-neural network-based algorithms, the neural network and method presented in this paper (e.g., Method 2800) showed a higher correlation with the number of all distinct transcripts per cell, thus indicating higher accuracy in identifying polymerase clones or clusters in flow cell images (e.g., transcripts) of in situ samples.

[0381] In some embodiments, methods 500, 2800 may include determining a bioanalyte based on base readings or classifications of polymerase clones in one or more sequencing cycles. The bioanalyte includes, but is not limited to, morphological features, transcripts, RNA, mRNA, proteins, or combinations thereof. For example, base readings or classifications of polymerase clones from six consecutive sequencing cycles of ATTCGA may indicate a cellular protein that can be uniquely barcoded with “ATTCGA”.

[0382] In some embodiments, method 2800 further includes the following operation (iv): in response to determining that the predicted classification of a first pixel among the one or more pixels is different from the background (e.g., the classification may include A, T, C, G, U, or background), determining a first morphological feature, a first RNA or mRNA, or a first protein based on the one or more predicted classifications. In some embodiments, method 2800 further includes the following operation (v): in response to determining that the predicted classification of a second pixel among the one or more pixels is different from the background classification (e.g., the classification may include A, T, C, G, U, or background), determining a second morphological feature, a second RNA or mRNA, or a second protein based on the one or more predicted classifications.

[0383] In some embodiments, method 2800 further includes the following operation (iv): determining a first morphological feature, a first RNA or mRNA, or a first protein based on a predicted base reading result of a first pixel in one or more cycles. In some embodiments, method 2800 further includes the following operation (v): determining a second morphological feature, a second RNA or mRNA, or a second protein based on a predicted base reading result of a second pixel in one or more cycles.

[0384] In some embodiments, method 2800 further includes determining the spatial relationship between a first pixel and a second pixel, the operation including one or more of the following: visualizing the first pixel and the second pixel in a common coordinate system; calculating the 2D or 3D spatial distance between the first pixel and the second pixel; and determining whether the first pixel and the second pixel are within the same polymerase clone.

[0385] In some embodiments, method 2800 further includes: (iv) in response to determining that the predicted classification of a first pixel among the one or more pixels differs from the background classification, determining at least a first target based on one or more predicted classifications: a first morphological feature; a first RNA or mRNA; and a first protein; and (v) in response to determining that the predicted classification of a second pixel among the one or more pixels differs from the background classification, determining at least one second target different from the first target based on one or more predicted classifications: a first morphological feature; a first RNA or mRNA; and a first protein. In some embodiments, the second target and the first target belong to different types of targets (e.g., protein and morphological feature), thereby advantageously enabling multi-omics analysis and research of the target bioanalyte using the methods herein. In some embodiments, the first target and the second target correspond to the bioanalyte of the sample. In some embodiments, method 2800 further includes: spatially aligning the positions of the first target and the second target based on the one or more predicted classifications; and determining the bioanalyte of the sample fixed on a support based on the spatial alignment.

[0386] In some embodiments, the method 2800 herein advantageously allows the use of the neural networks disclosed herein to achieve spatial alignment of two or more different bioanalytes, or in other words, to achieve their co-localization. Such different bioanalytes may belong to different types. For example, the first bioanalyte may be a morphological feature, and the second bioanalyte may be a protein or mRNA. Such different bioanalytes may be sequenced within the same sequencing run, in the same or different sequencing cycles. Exemplary embodiments of staining and sequencing different target analytes within cells or tissues are disclosed in PCT application PCT / US2025 / 10310, filed January 3, 2025, the entire contents of which are incorporated herein by reference. The number of different bioanalytes may be limited by the availability of a unique barcode that can be used to distinguish one bioanalyte from the others. For example, the number of different bioanalytes may range from 2 to 100, 4 to 350, 10 to 500, 50 to 1000, or more. For example, based on intensity prediction, base interpretation, and / or classification performed in one or more cycles using Method 500 or 2800, protein A can be located within the nucleus of a specific cell type, while protein B can be located adjacent to a transcript within the mitochondria but not within the cytosol. Identifying such distinct bioanalytes can advantageously provide more information, such as spatial relationships, which may aid in biological, physiological, or pathological analysis of samples being sequenced. In some embodiments, the bioanalytes described herein can be any physical feature of the sample or its source. The detection, localization, and spatial alignment of bioanalytes can correspond to a variety of physiological, biological, and pathological features of cells or tissues, which can advantageously provide information that may contribute to a better understanding of cellular function, regulation, and interactions, thereby advancing existing biomedical research, including but not limited to more effective disease modeling and drug discovery efforts.

[0387] In some embodiments, method 2800 further includes the following operations: (iv) determining the position of one or more of the following based on one or more predicted base interpretation results or corresponding positions of predicted classifications: a first morphological feature, a first RNA or mRNA, a first transcript, and a first protein. In some embodiments, method 2800 further includes the following operations: (v) determining the position of one or more of the following based on one or more second predicted base interpretation results or corresponding positions of predicted classifications: a second morphological feature, a second RNA or mRNA, a second transcript, and a second protein. In some embodiments, method 2800 further includes the following operations: (vi) spatially aligning the position of one or more of the second morphological feature, the second RNA or mRNA, and the second protein with the position of one or more of the first morphological feature, the first RNA or mRNA, and the first protein; and the following operation: (vii) determining the biomarkers of the sample fixed on the support based on the spatial alignment.

[0388] In some embodiments, method 2800 may include saving the base reading results obtained in operation 2812 in a predetermined format (e.g., a FastQ file compatible with subsequent operations) to enable subsequent analyses, such as linker trimming and secondary analysis.

[0389] Image patches from the flow cell image are used to predict base interpretation results.

[0390] In some embodiments, method 2800 may include operation 2812: (iii) performing a corresponding base readout for each of the determined polymerase clones by a processor. In some embodiments, operation 2812 includes extracting a plurality of image patches from a second plurality of flow cell images based on a polymerase clone map. The polymerase clone map may be generated using various algorithms, such as those from operation 2804 or 2804'. In some embodiments, operation 2812 further includes: providing input to a neural network, the input including the plurality of image patches, wherein each image patch includes one or more image patch images from a plurality of color channels, and wherein each image patch includes at least a portion of the second plurality of flow cell images; and using the neural network and based on the input to predict a plurality of base readout results, wherein each base readout result corresponds to a corresponding image patch.

[0391] In some embodiments, each corresponding image patch includes a polymerase clone located at or near the center of the corresponding image patch. For example, the distance between the polymerase clone and the center of the corresponding image patch may be no more than 1 to 10 pixels. In some embodiments, each image patch contains 3 to 128 pixels along a spatial dimension (e.g., along the x-direction or y-direction). The size of the image patch remains relatively small compared to the size of the flow cell image, for example, 10, 20, 50, 100, 500, 1000 times smaller than the size of the flow cell image, or less. In some embodiments, the plurality of image patches includes 100 to 10 8 Image blocks. In some embodiments, two or more distinct image blocks may at least partially overlap each other. In some embodiments, each image block may contain more than one, two, three, five, or ten polymerase clones, but only pixels of a single polymerase clone at its center are used to generate a base interpretation result corresponding to the image block. For example, when each image block comprises an image block image of size 32 by 32, the first image block may contain pixels 1 to 32 in both the x and y directions to cover the polymerase clone centered at pixel (16, 16) of the flow cell image, the second image block may contain pixels 2 to 33 in both the x and y directions to cover the second polymerase clone centered at pixel (17, 17.5), and the third image block may contain pixels 5 to 36 in both the x and y directions to cover the third polymerase clone centered at pixel (19, 19) of the flow cell image. In some embodiments, a very limited number of polymerase clones may be used in each image block, rather than using only a single polymerase clone, to generate a reference base interpretation result, reference strength, or prediction. The very limited number of polymerase clones can range from 1 to 4, 1 to 8, 1 to 20, 1 to 50, or 1 to 100. This very limited number of polymerase clones may be 100 times, 1000 times, or even 10 times less than the total number of polymerase clones in the corresponding flow cell image. 4 times, 10 5 times, 10 6 times, 10 7 times or 10 8 times.

[0392] In some embodiments, the number of pixels within each image patch can be optimized to balance computational complexity with the spatial context information to be included for training the neural network. The number of image patch images within each image patch can be optimized to balance computational complexity with the spatial context information within each image patch for accurate and reliable predictions using the neural network. In some embodiments, the number of pixels within each image patch can be based at least in part on the polymerase clone density of the sample being imaged. In some embodiments, each image patch can contain multiple pixels, but predictions can be performed only for a single polymerase clone located at or near the center of the image patch. Similarly, when training a neural network to predict base interpretation results, for example using method 2900, the reference base interpretation result is only for a single polymerase clone at or near the center of the image patch. In some embodiments, a very limited number of polymerase clones, rather than a single polymerase clone, can be used in each image patch to train the neural network or make predictions. This very limited number of polymerase clones can range from 1 to 4, 1 to 8, 1 to 20, 1 to 50, or 1 to 100. This very limited number of polymerase clones may be 100 times, 1000 times, or 10 times less than the total number of polymerase clones in the corresponding flow cell image. 4 times, 10 5 times, 10 6 times, 10 7 times or 10 8 times.

[0393] In some embodiments, each image block may include multiple image block images corresponding to different color channels. For example, each image block may include image block images covering the same pixels in three different color channels in the xy plane. The same pixels may be pixels determined after registration to correct for spatial offsets across different color channels. In some embodiments, each image block may include multiple image block images corresponding to different cycles (e.g., consecutive cycles n-1, n, n+1 in a sequencing run). For example, each image block may include 3 images, each from different color channels in 4 adjacent cycles, such that each image block may include a total of 12 image block images. When the sample is a 3D sample (e.g., an in situ cell sample), each image block may include 5 different z-levels, so that the total number of image block images is 60.

[0394] In some embodiments, at least two of the plurality of image blocks include at least partially overlapping image block images that contain some of the same pixels. In some embodiments, each of the plurality of image blocks includes pixels that at least partially overlap with another image block of the plurality of image blocks.

[0395] In some embodiments, the first plurality of flow cell images are acquired from only a single color channel, such that flow cell images acquired from different color channels may require different neural networks to predict high-resolution intensity, base interpretation results, classification, etc., as disclosed herein.

[0396] In some embodiments, the first plurality of flow cell images are acquired from only a single z-level, such that flow cell images acquired at different z-levels of a 3D sample (e.g., in situ cells) may require different neural networks to predict high-resolution intensity, base interpretation results, classification, etc., as disclosed herein.

[0397] In some embodiments, the first plurality of flow cell images are acquired from one or more cycles. In some embodiments, the one or more cycles include multiple cycles in a sequencing run. In some embodiments, the one or more cycles include a current cycle N, and the first plurality of flow cell images are acquired from at least one cycle preceding the current cycle N. The current cycle N is the cycle in the sequencing cycle in which sequencing is being performed. In some embodiments, flow cell images may have been acquired in the current cycle N, but no flow cell images may have been acquired in the next cycle N+1.

[0398] In some embodiments, operation 2802 (ii) of providing a second plurality of flow cell images as input to a neural network via a processor or a first reconfigurable logic device includes: (ii) providing the second plurality of flow cell images as input to the neural network via a processor or a first reconfigurable logic device, without providing the neural network with a polymerase clone map or polymerase clone positions in the second plurality of flow cell images as input. In other words, operation (ii) of method 2800 does not require providing the neural network with inputs such as a polymerase clone map, a list of polymerase clone positions, etc., to predict base readings. In some embodiments, the spatial location of polymerase clones within the flow cell images (e.g., the second plurality of flow cell images) is not used to predict base readings using a neural network. Instead, each image patch may contain relative spatial information of the polymerase clones relative to the remaining pixels in the same image patch, which can be used to predict base readings using a neural network. Method 2800 may predict base readings, for example, in operation 2812, without using inputs such as a polymerase clone map, a list of polymerase clone positions, etc. Instead, multiple image patches may be extracted from the second plurality of flow cell images using polymerase clone maps, lists of polymerase clone positions, etc.

[0399] In some embodiments, the operation of using a neural network to predict multiple base interpretations (where each base interpretation corresponds to a corresponding image patch) based on input includes: predicting a probability map for each of a plurality of color channels corresponding to the corresponding image patch; and determining the base interpretation for the corresponding image patch based on the probability map. For example, for a flow cell image from four different color channels, four different probability maps may be generated. The size and dimensions of each probability map may be the same as the flow cell image, or cover at least a portion of the flow cell image. Each pixel in the probability map may have a probability value corresponding to a channel. As an example, pixel (12,12) may have probability values ​​of 0.2, 0.01, 0.2, and 0.59 in the four different channels representing nucleotides A, T, C, and G, and the base interpretation for pixel (12,12) may be determined to be the highest probability among the different color channel probabilities, i.e., 0.59, and its base interpretation corresponds to nucleotide G. The neural network can be trained to predict the probability maps. In some embodiments, training a neural network to predict probability maps may be based on a reference polymerase clone map or any equivalent information indicating the location of polymerase clones, such as a list of polymerase clone locations. In some embodiments, the neural network for predicting probability maps may be trained by comparing each probability map to a corresponding reference polymerase clone map. In some embodiments, the neural network may be trained to minimize a loss function based on the comparison of the probability map to the corresponding reference polymerase clone map. For example, the probability maps may be initialized to have random values ​​in each pixel, and the neural network may be trained to produce higher values ​​for pixels in the probability maps corresponding to polymerase clones than for pixels corresponding to non-polymerase clone structures. In some embodiments, the sum of the values ​​of each pixel in all probability maps for different color channels may be summed to a fixed value, such as 1, 10, 100, etc. As an example, the values ​​of pixels (24, 25) in three probability maps corresponding to three different color channels may be 0.24, 0.51, and 0.25, which are summed to 1. In some embodiments, each base interpretation result corresponds to a corresponding image patch, which includes one or more image patch images. In some embodiments, the operation of using a neural network to predict the plurality of base interpretation results based on input includes: generating a first single intensity for a first channel among a plurality of color channels corresponding to a corresponding image patch; and determining the base interpretation result of the corresponding image patch based on the single intensity. As an example, the first single intensity of the first color channel can be determined using a prediction performed by the neural network disclosed herein. The first single intensity may be normalized or unnormalized. The first single intensity may correspond to a single polymerase clone of the corresponding image patch, the single polymerase clone containing one or more image patch images of the same polymerase clone in adjacent cycles of a sequencing run. The first single intensity may correspond to one of the adjacent cycles, for example, the current cycle.Base interpretation results can be determined based on the first single intensity of the current cycle, for example, by comparing the first single intensity with other intensities from the same polymerase clone from other color channels. These other intensities can be predicted in a similar manner using the same or different neural networks.

[0400] In some embodiments, the method further includes the following operations: using a second neural network to predict a second single intensity of a second channel among a plurality of color channels corresponding to a corresponding image patch; and determining the base reading result of the corresponding image patch based on at least a first single intensity and a second single intensity.

[0401] In some embodiments, the method further includes the following operations: using a second neural network or the same first neural network to predict a second single intensity of a second channel among a plurality of color channels corresponding to a corresponding image patch; and using a third neural network or the same first neural network to predict a third single intensity of a third channel among a plurality of color channels corresponding to a corresponding image patch; and determining a base interpretation result for the corresponding image patch based on at least the first single intensity, the second single intensity, and the third single intensity. For example, in the current cycle N, for the same polymerase clone, different neural networks (e.g., each neural network can be trained with different training data but has the same number of neural network layers and parameters) can be used to predict the first intensity, the second intensity, and the third intensity as 50, 690, and 80. The base interpretation result of the polymerase clone may correspond to a nucleotide with an intensity of 690 in the second color channel, but no nucleotide with this intensity in the first or third color channel.

[0402] In some embodiments, the operation (iii) of using a neural network to predict one or more base interpretation results corresponding to one or more polymerase clones in a second plurality of flow cell images via a first reconfigurable device or integrated circuit includes: identifying two or more pixels in the second plurality of flow cell images as repeats of a single polymerase clone; and selecting one of the two or more pixels as the center of the single polymerase clone. In some embodiments, the two or more pixels may be at the same z-level. In some embodiments, the two or more pixels may be at different z-levels. Exemplary embodiments of the operation of identifying two or more pixels in the second plurality of flow cell images as repeats of a single polymerase clone and selecting one of the two or more pixels as the center of the single polymerase clone are disclosed in PCT application PCT / US23 / 76125, the entire contents of which are incorporated herein by reference.

[0403] While the embodiments described herein focus on using and training neural networks, other AI-based models can also be used for similar purposes. In some embodiments, methods 500 and 2800 of this document can be performed using AI-based models other than neural networks. In some embodiments, methods 600, 700, and 2900 can be used to train AI-based models other than neural networks to make predictions or inferences using method 500 or 2800. Some non-limiting examples of AI-based models include: random forests, decision trees, k-means clustering, and gradient boosting trees. In some embodiments, AI-based models can be used to predict intensity, classification, or base interpretation results by processing intensity from flow cell images and / or high-resolution flow cell images. In some embodiments, AI-based models other than neural networks can use information including only intensity to predict intensity, classification, or base interpretation results, and such information may lack the spatial context of the intensity, the shape of the polymerase clone, background noise, signals from other cellular structures, etc. In some embodiments, the neural network of this paper predicts intensity, classification, or base interpretation results by advantageously using flow cell images or high-resolution flow cell images that include not only intensity information but also other information, including but not limited to background noise, polymerase clone size and shape, spatial relationships between polymerase clones, etc., to achieve more accurate predictions or inferences.

[0404] In some embodiments, the neural network described herein is a convolutional neural network (CNN). In some embodiments, the neural network is a 3D CNN. In some embodiments, the neural network is a 2D CNN. In some embodiments, the neural network includes one or more convolutional layers. In some embodiments, the neural network is a recurrent neural network (RNN). In some embodiments, the neural network is a 3D RNN. In some embodiments, the neural network is a 2D RNN. In some embodiments, the neural network includes one or more long short-term memory (LSTM) layers. In some embodiments, the neural network is a U-Net. In some embodiments, the neural network includes a residual network (ResNet). In some embodiments, the neural network may include a Transformer-based model, such as the visual Transformer (ViT). In some embodiments, the neural network includes a U-Net having a first predetermined repetition of downsampling and convolution operations and a subsequent second predetermined repetition of upsampling, concatenation, and convolution operations. The first predetermined repetition and the second predetermined repetition may have the same number, for example, 3 or 4 times. In some embodiments, the neural network is a U-Net having a first predetermined number of filters in each downsampling repetition and then a second predetermined number of filters in each upsampling and / or concatenation repetition. For example, in three repetitions, the first predetermined number of filters can be 32, 64, 128, and 256 filters, while in the corresponding three repetitions, the second predetermined number of filters can be 128, 64, 64, and 32 filters, respectively. As another example, in three repetitions, the first predetermined number of filters can be 32, 64, 128, and 256 filters, while in the corresponding three repetitions, the second predetermined number of filters can be 256, 128, 64, and 32 filters, respectively.

[0405] In some embodiments, ...

Claims

1. A computer-implemented method, comprising: (i) A first plurality of flow cell images of a sample immobilized on a support are generated by a sequencing system through one or more cycles of sequencing reactions in one or more color channels, wherein the first plurality of flow cell images are acquired at a first resolution; (ii) The first plurality of flow pool images are provided as input to the neural network via a processor or a first reconfigurable logic device, wherein the neural network is pre-trained using a training dataset of training flow pool images. (iii) Using the neural network via a first reconfigurable device or integrated circuit to predict a second plurality of flow pool images, each of the second plurality of flow pool images having a second resolution and corresponding to a corresponding image in the first plurality of flow pool images, and wherein the second resolution is at least 2 to 32 times larger than the first resolution in one or more spatial dimensions. (iv) Determining polymerase clones from the second plurality of flow cell images using the processor, the first reconfigurable logic device, or the integrated circuit; as well as (v) Performing a corresponding base reading on each of the identified polymerase clones based on the second plurality of flow cell images, using the processor, the first reconfigurable logic device, or the integrated circuit.

2. A computer-implemented method for predicting base identification results, comprising: (i) Using a sequencing system, by performing one or more cycles of sequencing reactions in one or more color channels, a first plurality of flow cell images of a sample immobilized on a support are generated, the first plurality of flow cell images including a first resolution; (ia) Generate a second plurality of flow pool images including a second resolution by means of a processor, a first reconfigurable logic device or an integrated circuit, wherein the second resolution is at least 2 to 32 times larger than the first resolution in one or more spatial dimensions; (ii) The second plurality of flow cell images are provided as input to the neural network via the processor, the first reconfigurable logic device, or the integrated circuit; or The processor, the first reconfigurable logic device, or the integrated circuit provides the second plurality of flow cell images to the polymerase clone map generation algorithm or the base determination algorithm. (iii) Using the neural network via the first reconfigurable device or the integrated circuit, predict one or more base interpretation results corresponding to one or more polymerase clones of the second plurality of flow cell images; or The neural network is used, via the first reconfigurable device or the integrated circuit, to predict one or more classifications corresponding to one or more pixels in the second plurality of flow pool images.

3. The computer-implemented method according to claim 2, further comprising: (iia) The polymerase clone map is determined based on the second plurality of flow cell images using the processor, the first reconfigurable device, or the integrated circuit; as well as (iiia) Using the processor, the first reconfigurable logic device, or the integrated circuit, determine the corresponding position of one or more predicted base interpretation results or one or more predicted classifications based on the polymerase clone map.

4. The computer-implemented method of claim 2, wherein the one or more pixels include at least one pixel not included in any polymerase clone of the polymerase clone map.

5. The computer-implemented method of claim 2, wherein the one or more pixels include at least one pixel not included in any polymerase clone in the polymerase clone diagram and at least one pixel included in at least one polymerase clone in the polymerase clone diagram.

6. The computer-implemented method according to claim 2, further comprising: (iv) In response to determining that a first pixel among the one or more pixels has a predicted classification different from the background classification, determining a first morphological feature, a first RNA or mRNA, or a first protein based on the one or more predicted classifications; and (v) In response to determining that a second pixel among the one or more pixels has a predicted classification different from the background classification, a second morphological feature, a second RNA or mRNA, or a second protein is determined based on the one or more predicted classifications.

7. The computer-implemented method according to claim 3, further comprising: (iv) Determine the position of one or more of the following based on the corresponding position of the one or more predicted base interpretation results or predicted classification: first morphological feature, first RNA or mRNA, and first protein.

8. The computer-implemented method according to claim 7, further comprising: (v) Determine the position of one or more of the following based on the corresponding position of one or more second predicted base interpretation results or predicted classifications: second morphological feature, second RNA or mRNA, and second protein.

9. The computer-implemented method according to claim 8, further comprising: (vi) Spatially align the positions of one or more of the second morphological feature, the second RNA or mRNA, and the second protein with the positions of one or more of the first morphological feature, the first RNA or mRNA, and the first protein. as well as (vii) Determine the biological characteristics of the sample fixed on the support based on spatial alignment.

10. The computer-implemented method according to any one of claims 2 to 9, wherein the neural network is pre-trained using a training dataset of training flow pool images.

11. The computer-implemented method of claim 10, wherein the neural network is pre-trained using a training dataset of training flow pool images prior to operation (iia).

12. The computer-implemented method according to any one of the preceding claims, wherein the neural network is not trained or retrained after operation (iia) and before operation (iii).

13. The computer-implemented method according to any one of the preceding claims, wherein the sample contains polymolecules therein, and the first plurality of flow cell images are acquired at z-stacks at different z-positions, having the first resolution along the z-direction.

14. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images are acquired from a single color channel.

15. The computer-implemented method according to any one of the preceding claims, wherein the one or more loops include a current loop N, and the first plurality of flow pool images are acquired from at least one loop preceding the current loop N.

16. The computer-implemented method according to any one of the preceding claims, wherein the one or more cycles comprise multiple cycles in a sequencing run.

17. A computer-implemented method according to any one of the preceding claims, wherein (iia) determining the polymerase clone map based on the second plurality of flow cell images via the first reconfigurable device or the integrated circuit comprises: The polymerase clone map is predicted based on the second plurality of flow cell images using the neural network and via the first reconfigurable device or the integrated circuit.

18. The computer-implemented method according to any one of the preceding claims, wherein (iia) comprises: Using the first reconfigurable device or the integrated circuit, the neural network is used to predict the base reading result of each polymerase clone corresponding to the second plurality of flow cell images at the second or third resolution. as well as The polymerase clone map is determined based on the predicted base interpretation results at the second or third resolution and the corresponding quality index of each predicted base interpretation result.

19. The computer-implemented method according to any one of the preceding claims, wherein the third resolution is at least 2 to 32 times larger than the first resolution or the second resolution in one or more spatial dimensions.

20. The computer-implemented method according to any one of the preceding claims, wherein the third resolution is greater than the first resolution and the second resolution in one or more spatial dimensions.

21. The computer-implemented method according to any one of the preceding claims, wherein the polymerase clone map includes the spatial coordinates of each of at least some of the polymerase clones in the second plurality of flow cell images.

22. The computer-implemented method according to any one of claims 2 to 21, wherein (ii) providing the second plurality of flow pool images as input to the neural network via the processor or the first reconfigurable logic device comprises: (ii) The processor or the first reconfigurable logic device provides the second plurality of flow cell images as input to the neural network, without providing the polymerase clone image or the position of the polymerase clone in the second plurality of flow cell images as input to the neural network.

23. The computer-implemented method according to any one of claims 2 to 22, wherein (iii) using the neural network to predict base interpretation results of one or more polymerase clones corresponding to the second plurality of flow cell images via the first reconfigurable device or the integrated circuit comprises: Multiple image patches are extracted from the second plurality of flow cell images based on the polymerase clone map; The neural network is provided with input, the input comprising the plurality of image patches, wherein each image patch comprises one or more image patch images from the one or more color channels, and wherein each image patch comprises at least a portion of the second plurality of flow pool images; as well as The neural network is used to predict multiple base interpretation results based on the input, where each base interpretation result corresponds to a corresponding image patch.

24. The computer-implemented method of claim 23, wherein each corresponding image block includes a polymerase clone located at or adjacent to the center of the corresponding image block.

25. The computer-implemented method according to any one of claims 23 to 24, wherein each corresponding image block comprises 2 to 128 pixels along the spatial dimension.

26. The computer-implemented method according to any one of claims 23 to 25, wherein the plurality of image blocks comprises 100 to 10 6 Image blocks.

27. The computer-implemented method according to any one of claims 23 to 26, wherein at least two of the plurality of image blocks comprise at least partially overlapping image block images, the at least partially overlapping image block images comprising the same pixels.

28. The computer-implemented method according to any one of claims 23 to 27, wherein each of the plurality of image blocks includes pixels that at least partially overlap with another image block of the plurality of image blocks.

29. The computer-implemented method according to any one of the preceding claims, wherein the one or more color channels comprise 2, 3, or 4 color channels.

30. The computer-implemented method according to any one of claims 2 to 29, the method further comprising: The first plurality of flow cell images are upsampled to generate the second plurality of flow cell images.

31. The computer-implemented method according to any one of claims 23 to 30, wherein each image block comprises one or more image block images from the one or more color channels and one or more cycles.

32. The computer-implemented method according to any one of the preceding claims, wherein the one or more cycles comprise a continuous cycle of sequencing runs.

33. The computer-implemented method according to any one of claims 23 to 32, wherein the neural network is used and the plurality of base interpretation results are predicted based on the input, wherein each base interpretation result corresponds to a corresponding image patch, comprising: A probability map is predicted for each of the one or more color channels corresponding to the corresponding image block, wherein each probability map includes a probability value for base interpretation for each pixel of the corresponding image block; as well as The base interpretation result of the corresponding image block is determined based on the probability map of the one or more channels.

34. The computer-implemented method according to any one of claims 23 to 33, wherein the neural network is used and the plurality of base interpretation results are predicted based on the input, wherein each base interpretation result corresponds to a corresponding image patch, comprising: A first single intensity is generated for a first channel among the one or more color channels corresponding to the corresponding image patch; as well as The base interpretation result of the corresponding image block is determined based on the single intensity.

35. The computer-implemented method according to any one of claims 23 to 34, wherein the neural network is used and the plurality of base interpretation results are predicted based on the input, wherein each base interpretation result corresponds to a corresponding image patch, comprising: A first single intensity is generated for a first channel among the one or more color channels corresponding to the corresponding image patch; as well as The base interpretation result of the corresponding image block is determined based on the single intensity.

36. The computer-implemented method according to claim 35, further comprising: A second neural network is used to predict a second single intensity for a second channel among the one or more color channels corresponding to the corresponding image patch; as well as The base interpretation result of the corresponding image block is determined based on at least the first single intensity and the second single intensity.

37. The computer-implemented method according to claim 36, further comprising: A second neural network is used to predict a second single intensity for a second channel among the one or more color channels corresponding to the corresponding image patch; as well as A third single intensity is predicted using a third neural network for the third channel among the one or more color channels corresponding to the corresponding image patch; as well as The base interpretation result of the corresponding image block is determined based on at least the first single intensity, the second single intensity, and the third single intensity.

38. The method of claim 2, wherein (iii) using the neural network via the first reconfigurable device or the integrated circuit to predict one or more base interpretation results corresponding to one or more polymerase clones of the second plurality of flow cell images comprises: Identify two or more pixels in the second plurality of flow cell images as replicates of a single polymerase clone; as well as One of the two or more pixels is selected as the center of the single polymerase clone.

39. The computer-implemented method of claim 1, wherein the sample contains polymolecules therein, and wherein the first plurality of flow cell images are acquired at z-stacks at different z-positions.

40. A computer-implemented method, comprising: (i) By using a sequencing system, through one or more cycles of performing a sequencing reaction, a first plurality of flow cell images of a sample immobilized on a support are generated, wherein the first plurality of flow cell images have a first resolution; (ii) The first plurality of flow cell images are provided as input to the neural network via a processor or a first reconfigurable logic device, wherein the neural network is pre-trained using a training dataset of training flow cell images and reference base interpretation results of the training dataset. (iii) Using the neural network via a first reconfigurable device or integrated circuit to predict a second plurality of flow pool images, each of the second plurality of flow pool images having a second resolution and corresponding to a corresponding image in the first plurality of flow pool images, and wherein the second resolution is at least 2 to 32 times larger than the first resolution in one or more spatial dimensions. (iv) Determining polymerase clones from the second plurality of flow cell images using the processor, the first reconfigurable logic device, or the integrated circuit; as well as (v) Performing a corresponding base reading on each of the identified polymerase clones based on the second plurality of flow cell images, using the processor, the first reconfigurable logic device, or the integrated circuit.

41. The method according to any one of the preceding claims, wherein the sample is a cell sample comprising in situ cells, tissue, and both.

42. The method according to any one of the preceding claims, wherein the neural network is a convolutional neural network.

43. The method according to any one of the preceding claims, wherein the sample is a 3D sample and the neural network is a 3D neural network.

44. The method according to any one of the preceding claims, wherein the sample is a 2D sample, and wherein the neural network is a 2D neural network.

45. The computer-implemented method according to any one of the preceding claims, wherein the neural network is a convolutional neural network.

46. ​​The computer-implemented method according to any one of the preceding claims, wherein the neural network is pre-trained using one or more loss functions based on comparing training base interpretation results of the training flow cell image with reference base interpretation results of the training flow cell image.

47. The computer-implemented method according to any one of the preceding claims, wherein the training flow cell image is derived from only a single color channel.

48. The computer-implemented method according to any one of the preceding claims, wherein each of the training flow cell images comprises flow cell images from the same field of view of multiple sequencing cycles stacked along the time dimension.

49. The computer-implemented method according to any one of the preceding claims, wherein each of the training flow pool images comprises flow pool images from the same field of view from one or more cycles.

50. The computer-implemented method according to any one of the preceding claims, wherein each of the training flow cell images includes a flow cell image of the sample at one or more z-levels.

51. The computer-implemented method according to any one of the preceding claims, wherein (iii) using the neural network to predict the second plurality of flow pool images includes predicting a high-resolution post-processed image of the first plurality of flow pool images, and wherein the processing includes one or more of the following: noise reduction, background reduction; intensity shift correction; intensity normalization; color correction; hysteresis and / or out-of-step; image registration; and deconvolution.

52. The computer-implemented method according to any one of the preceding claims, wherein (v) performing corresponding base readings on each of the determined polymerase clones based on the second plurality of flow cell images does not use a neural network.

53. The computer-implemented method according to any one of the preceding claims, wherein (iv) determining the polymerase clone from the second plurality of flow cell images does not use a neural network.

54. The computer-implemented method according to any one of the preceding claims, wherein (iv) determining the polymerase clone from the second plurality of flow cell images comprises determining the location of the center of each of the polymerase clones from the second plurality of flow cell images.

55. The computer-implemented method according to any one of the preceding claims, wherein (iv) determining the polymerase clone from the second plurality of flow cell images does not use a neural network.

56. The computer-implemented method according to any one of the preceding claims, wherein the neural network comprises a first image processing part and a second base determination part.

57. A computer-implemented method according to any one of the preceding claims, wherein predicting the second plurality of flow cell images using the neural network via the first reconfigurable device or the integrated circuit comprises: The output image is generated from the first image processing part of the neural network as the second plurality of flow cell images, without passing through the second base determination part of the neural network.

58. The computer-implemented method according to any one of the preceding claims, wherein the first image processing portion comprises at least a portion of one or more of the following: an input layer, a convolutional layer, a pooling layer, an embedding layer, an output layer, and an encoder of the neural network, and wherein the second base interpretation portion lacks any one of the following: a convolutional layer, a pooling layer, an embedding layer, an output layer, an encoder, and a decoder of the neural network.

59. The computer-implemented method according to any one of the preceding claims, wherein the first image processing portion comprises at least a portion of one or more of the following: an input layer, a convolutional layer, a pooling layer, an embedding layer, and an encoder of the neural network, and wherein the second base interpretation portion comprises at least a portion of one or more of the following: an input layer, an output layer, a convolutional layer, a pooling layer, an embedding layer, an encoder, and a decoder of the neural network.

60. A computer-implemented method according to any one of the preceding claims, wherein the processor comprises one or more of the following: CPU, GPU, TPU, NPU, FPGA, and AI chip.

61. A computer-implemented method according to any one of the preceding claims, wherein the processor comprises one or more of the following: CPU, NPU, and FPGA.

62. The computer-implemented method according to any one of the preceding claims, wherein the processor lacks any NPU, FPGA, or AI chip.

63. The computer-implemented method according to any one of the preceding claims, wherein the first reconfigurable logic device comprises one or more FPGA units.

64. The computer-implemented method according to any one of the preceding claims, wherein the integrated circuit comprises one or more NPUs, AI chips, or both.

65. A computer-implemented method according to any one of the preceding claims, wherein one or more of the processor, the first reconfigurable logic device, and the integrated circuit are included in a single housing of the sequencing system.

66. A computer-implemented method according to any one of the preceding claims, wherein the integrated circuit communicates data with the first reconfigurable logic device.

67. The computer-implemented method according to any one of the preceding claims, wherein the convolutional network includes U-Net.

68. The computer-implemented method according to any one of the preceding claims, wherein the first convolution comprises a 3D convolution having a kernel.

69. The sequencing system according to any one of the preceding claims, wherein the first convolution comprises a 2D convolution having a kernel.

70. The sequencing system according to any one of the preceding claims, wherein the convolutional kernel has at least three dimensions.

71. The sequencing system according to any one of the preceding claims, wherein the neural network is a 3D convolutional neural network.

72. The sequencing system according to any one of the preceding claims, wherein the neural network is a 2D convolutional neural network.

73. The computer-implemented method according to any one of the preceding claims, wherein the first resolution and the second resolution are 3D resolutions.

74. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images are derived from a single color channel.

75. The computer-implemented method according to any one of the preceding claims, wherein (v) performing a corresponding base reading on each of the determined polymerase clones based on the second plurality of flow cell images by the processor comprises: The processor performs corresponding base readings on each of the identified polymerase clones based on a fourth plurality of flow cell images, wherein the fourth plurality of images are predicted using a second neural network based on a third plurality of flow cell images.

76. The computer-implemented method according to any one of the preceding claims, wherein the third plurality of flow cell images are acquired from one or more color channels different from the single channel, and wherein the third plurality of flow cell images include the first resolution.

77. The computer-implemented method according to any one of the preceding claims, wherein the fourth plurality of flow cell images includes the second resolution.

78. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images are derived from one or more color channels.

79. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images comprises: In one or more cycles, the unbalanced diversity of nucleotide bases A, G, C, and T / U in the polynucleotide molecules immobilized on the support.

80. The computer-implemented method according to any one of the preceding claims, wherein two or more different multiply molecules in the multiply have different insertion sequences corresponding to different target RNA molecules or target cDNA molecules.

81. The computer-implemented method according to any one of the preceding claims, wherein each position of the determined polymerase clone corresponds to a position of the polynucleotide molecule.

82. The computer-implemented method according to any one of the preceding claims, wherein the unbalanced diversity of nucleotide bases A, G, C, and T / U in the multiply molecule comprises: In one or more cycles, (1) the percentage of the number of one or more types of nucleotide bases to the total number of (2) bases is less than 20%, 15%, 10% or 5%.

83. The computer-implemented method according to any one of the preceding claims, wherein the sample comprises a spatial density ranging from 10 3 Up to 10 10 Each mm 2 Overloaded multiply molecules.

84. The computer-implemented method according to any one of the preceding claims, wherein the first resolution ranges from 0.1 μm to 5 μm.

85. The computer-implemented method according to any one of the preceding claims, wherein the downsampling factor is 2, 4 or 8.

86. The computer-implemented method according to any one of the preceding claims, wherein one or more of operations (ii) to (v) are performed while a sequencing run is being executed.

87. The computer-implemented method according to any one of the preceding claims, wherein the one or more loops include the current loop N.

88. The computer-implemented method according to any one of the preceding claims, wherein N ranges from 1 to 500.

89. The computer-implemented method according to any one of the preceding claims, wherein one or more of operations (ii) to (v) are performed simultaneously in a cycle following the current cycle N, either before or during the current cycle N.

90. The computer-implemented method according to any one of the preceding claims, wherein the training dataset of flow cell images comprises a z-stack of flow cell images acquired at different z-positions.

91. The computer-implemented method according to any one of the preceding claims, wherein the second resolution is at least 4, 6, or 8 times larger than the first resolution in all three dimensions.

92. The computer-implemented method according to any one of the preceding claims, further comprising: Register the second plurality of flow cell images to a common coordinate system.

93. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images are acquired from a single color channel of the sequencing system.

94. A computer-implemented method according to any one of the preceding claims, wherein (vi) determining the polymerase clone from the second plurality of flow cell images by the processor comprises: Based on the identified polymerase clones described in (iv), a polymerase clone map including the spatial locations of the polymerase clones is generated.

95. The computer-implemented method according to any one of the preceding claims, wherein generating a polymerase clone map including the spatial locations of the polymerase clones based on the determined polymerase clones described in (iv) further comprises: Remove duplicate polymerase clones from the identified polymerase clones, wherein the duplicate polymerase clones are defocused.

96. A computer-implemented method according to any one of the preceding claims, wherein determining the polymerase clone from the second plurality of flow cell images by the processor comprises: Overlay the second plurality of flow cell images with the corresponding cell staining images; as well as The polymerase clone map is generated by including only polymerase clones located within the cell boundaries in the corresponding cell staining image.

97. The computer-implemented method according to any one of the preceding claims, wherein the support comprises a glass or plastic substrate.

98. The computer-implemented method according to any one of the preceding claims, wherein the support is included in a flow cell apparatus.

99. The computer-implemented method according to any one of the preceding claims, further comprising: The sample is provided containing multiple RNAs, including a first target RNA molecule and a second target RNA molecule.

100. The computer-implemented method according to any one of the preceding claims, further comprising: Multiple cDNA molecules are generated inside the sample, including a first target cDNA molecule corresponding to the first target RNA molecule and a second target cDNA molecule corresponding to the second target RNA molecule.

101. The computer-implemented method according to any one of the preceding claims, further comprising: The plurality of cDNA molecules in the sample are brought into contact with a plurality of target-specific locking probes, the plurality of target-specific locking probes including at least a first plurality of first target-specific locking probes and a second plurality of second target-specific locking probes.

102. The computer-implemented method according to any one of the preceding claims, further comprising: The plurality of RNA molecules in the sample are brought into contact with a plurality of target-specific locking probes, the plurality of target-specific locking probes including at least a first plurality of target-specific locking probes and a second plurality of target-specific locking probes.

103. The computer-implemented method according to any one of the preceding claims, wherein a single locking probe in the first plurality of first target-specific locking probes comprises: A first terminal region and a second terminal region, wherein the first terminal region selectively hybridizes with a first region of the first target cDNA molecule or the first target RNA molecule, and the second terminal region selectively hybridizes with a second region of the first target cDNA molecule or the first target RNA molecule.

104. The computer-implemented method according to any one of the preceding claims, wherein the first target-specific lock probe comprises a first target barcode sequence, the first target barcode sequence corresponding to and uniquely identifying the first target cDNA sequence or the first target RNA sequence.

105. The computer-implemented method according to any one of the preceding claims, wherein the first target-specific lock probe comprises a first target barcode sequence, the first target barcode sequence being positioned adjacent to one of the regions of the first target-specific lock probe that selectively hybridize with the first target cDNA molecule or the first target RNA sequence.

106. The computer-implemented method according to any one of the preceding claims, wherein the first target-specific lock probe comprises at least one universal connector sequence.

107. The computer-implemented method according to any one of the preceding claims, wherein the first target-specific lock probe comprises a universal primer binding site or a complementary sequence thereof for rolling circle amplification primers.

108. The computer-implemented method according to any one of the preceding claims, wherein the first target-specific locking probe comprises a universal compacted oligonucleotide binding site or its complementary sequence.

109. The computer-implemented method according to any one of the preceding claims, wherein the plurality of nucleotide reagents comprises: Multivalent molecules, nucleotides, nucleotide analogs, or combinations thereof.

110. The computer-implemented method according to any one of the preceding claims, wherein a single nucleotide or nucleotide analogue is detectably labeled or unlabeled.

111. The computer-implemented method according to any one of the preceding claims, wherein the detectably labeled single nucleotide or nucleotide analog comprises different detectable color markers corresponding to each different type of nucleotide base A, G, C, and T / U.

112. The computer-implemented method according to any one of the preceding claims, wherein a single multivalent molecule comprises a core with a plurality of nucleotide arms attached, and each arm of the single multivalent molecule comprises the same type of nucleotide base.

113. The computer-implemented method according to any one of the preceding claims, wherein generating the first plurality of flow cell images comprises: In each cycle, the optical color signal emitted from the nucleotide reagent bound to the plurality of polynucleotide molecules is imaged by an optical system.

114. The computer-implemented method according to any one of the preceding claims, wherein the first plurality of flow cell images comprises optical color signals emitted from the nucleotide reagents bound to the plurality of polymolecules.

115. The computer-implemented method according to any one of the preceding claims, further comprising: Remove the first sequencing read product from the first polypeptide molecule and retain the first polypeptide molecule in the sample, and The second sequencing read product is removed from the second polymolecule and the second polymolecule is retained in the sample.

116. The computer-implemented method according to any one of the preceding claims, further comprising: The plurality of polymers are sequenced repeatedly by repeating the following operation at least once: The first plurality of flow cell images of a sample immobilized on a support are generated by performing one or more cycles of sequencing reaction, thereby generating the first sequencing read product and the second sequencing product, wherein the sample contains a plurality of polyp molecules, wherein a first polyp molecule of the plurality of polyp molecules corresponds to a first target RNA molecule of the sample, and a second polyp molecule of the plurality of polyp molecules corresponds to a second target RNA molecule of the sample, wherein the first plurality of flow cell images. as well as The first sequencing read product is removed from the first polymolecule and the first polymolecule is retained in the sample, and the second sequencing read product is removed from the second polymolecule and the second polymolecule is retained in the sample.

117. The computer-implemented method according to any one of the preceding claims, wherein the first sequencing read product comprises some or all of the following: a first target barcode sequence in one or more tandem units of the first multiply molecule; a first insert sequence in one or more tandem units of the first multiply molecule; or a combination thereof.

118. The computer-implemented method according to any one of the preceding claims, further comprising: The presence of the first target RNA molecule, the second target RNA molecule, or both molecules in the sample is confirmed by base reading of the second plurality of flow cell images performed at the base reading positions in the base reading template.

119. The computer-implemented method according to any one of the preceding claims, further comprising: The sequencing system generates a second plurality of flow cell images of the sample immobilized on the support by performing subsequent cycles of sequencing reactions after one or more cycles.

120. A computer-implemented method according to any one of the preceding claims, wherein generating the first plurality of flow cell images of the sample fixed on the support comprises: Under conditions that inhibit sequencing of the second polyp, at least the first polyp within the sample is sequenced.

121. The computer-implemented method according to any one of the preceding claims, wherein sequencing at least the first polyp within the sample comprises: Multiple first sequencing reads are generated, and the sequences of the first sequencing reads are compared with a first target reference sequence to confirm the presence of the first target RNA in the sample.

122. The computer-implemented method according to any one of the preceding claims, wherein generating the first plurality of flow cell images of the sample fixed on the support comprises: Under conditions that inhibit sequencing of the first polyp, at least the second polyp within the sample is sequenced.

123. The computer-implemented method according to any one of the preceding claims, wherein sequencing at least the second polymorph within the cell sample comprises: Multiple second sequencing reads are generated, and the sequences of the second sequencing reads are compared with a second target reference sequence to confirm the presence of the second target RNA in the sample.

Images