Artificial intelligence cell sorting system and method

Abstract

Disclosed herein are devices, systems, related methods, computing devices, and computer-readable media relating to cell sorting by embedding using a real-time cell sorter. For example, in some embodiments, a method may include receiving first cell sorter data. A first cell sorter data meter may include cell sorter data including microscope data, hyperspectral imaging data, multispectral data, high dimensional vector data, or one or more combinations thereof. In some embodiments, the cell sorter data may include quantitative fluorescence data representative of one or more antibodies bound per cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecules (MESF), one or more other fluorescent quantitative indicators, or one or more combinations thereof. In some embodiments, the quantitative fluorescence data comprises one or more fluorescence signals: one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, or one or more combinations thereof.

Description

[0001] Cross Reference to Related Applications

[0002] This application claims the benefit of U.S. Provisional Application No. 63 / 510,959, filed June 29, 2023, the entire contents of which are incorporated herein by reference. BACKGROUND

[0003] Cell sorting generally involves sorting cells from a heterogeneous cell sample, but cell sorting presents many challenges. Current cell sorting strategies based on nonlinear embedding rely on algorithms that construct exclusion-gating sequences on two-dimensional density plots based on cell marker expression levels. These gates and sequences are subjective, can be imprecise, and do not leverage relationships between three or more cell markers on a single gate. Therefore, more sophisticated techniques are needed to sort cells and perform related analyses. The present invention addresses this need. BRIEF DESCRIPTION OF DRAWINGS

[0004] Embodiments will be readily understood by the following detailed description in conjunction with the accompanying drawings. For purposes of clarity, the same reference numbers will be used in the description and the drawings to identify the same or like components. In the drawings, related embodiments are shown in example, non- limiting detail through various views. The drawings include many features in detail for the purpose of clarity, but the purpose and operation of these features will be understood by one of ordinary skill in the art, and these features can take any suitable form. Any feature of the drawings can be used in combination with suitable features of other drawings, and / or in combination with any suitable feature of any embodiment of the present disclosure.

[0005] Figure 1A , 1B and 1C are example cell sorter support module block diagrams that perform support operations based on various embodiments.

[0006] Figure 2A , 2B and 2C are flowcharts of example methods that perform support operations based on various embodiments.

[0007] Figure 3 is an example graphical user interface that can be used to perform some or all of the support methods disclosed herein based on various embodiments.

[0008] Figure 4 is an example computing device block diagram that can be used to perform some or all of the support methods disclosed herein based on various embodiments.

[0009] Figure 5 is an example cell sorter support system block diagram that can be used to perform some or all of the cell sorting support methods disclosed herein based on various embodiments.

[0010] Figure 6A ,6B 6C is an exemplary workflow as contemplated in embodiments of this disclosure. Figure 6A The left image depicts a spectral cell sorter that records cells as events. The spectral cell sorter is capable of capturing detailed spectral features of each cell. The middle image is a graphical representation of the neural network used to process the spectral features. The right image depicts an exemplary gating system for parameterized embedding. Figure 6B An exemplary workflow for using a neural network trained to reproduce nonparametric embeddings is described. Figure 6C An exemplary workflow for using a neural network trained to directly generate nonparametric embeddings is described.

[0011] Figure 7A , 7B Figures 7A, 7B, and 7C depict a simple, representative gating process for sorting CD4+ T cell samples using a spectral cell sorter (from 7A to 7B to 7C).

[0012] Figure 8A and 8B Depicting in tSNE ( Figure 8A Embedding and UMAP ( Figure 8B The image of CD4+ T cells is highlighted in the embedding.

[0013] Figures 9A to 9C Depicting from embedding ( Figure 9A Exemplary choices for clustering in , and subsequent conventional gating sequences determined using gating localization algorithms as envisioned in this paper ( Figure 9B and 9C ).

[0014] Figures 10A to 10D A graphical representation depicting an exemplary mapping process as contemplated in this disclosure is provided. Figure 10A and 10B The workflow for reproducing nonparametric embeddings using a neural network is shown. Figure 10C and 10D The workflow for generating embeddings directly using a trained network is shown. Figure 10A The figure above illustrates the same gating shown in the initial nonparametric UMAP embedding, with neural network scoring of training data including 9901 events, and with neural network scoring of the remaining 255,975 events. Figure 10C The top figure illustrates the gating achieved by scoring 9901 events and the remaining 255,975 events using a neural network. The bottom figure shows a box plot illustrating the similarity of the median and percentile ranges of the above gating across different embeddings. Detailed Implementation

[0015] This document discloses devices, systems, methods, computing devices, and computer-readable media related to cell sorting using a real-time cell sorter via embedding. For example, in some embodiments, the method may include receiving first cell sorter data. The first cell sorter data may include cell sorter data. The cell sorter data may include fluorescence data, light scattering data, and / or a combination of both. The first cell sorter data may also include microscopy data, hyperspectral imaging data, multispectral data, high-dimensional vector data, or one or more combinations of the above. In some embodiments, the cell sorter data may include quantitative fluorescence data. The quantitative fluorescence data may include fluorescence intensity values ​​associated with one or more fluorescent dyes or fluorophores. Each of the one or more fluorescent dyes or fluorophores may be associated with a cell or protein marker associated with one or more cell populations. The one or more fluorescent dyes or fluorophores may be bound to or conjugated to one or more antibodies. The quantitative fluorescence data may be presented in the form of a two-dimensional or three-dimensional linear plot, a two-dimensional or three-dimensional logarithmic plot, a two-dimensional or three-dimensional bi-exponential plot, or one or more combinations of the above forms. Quantitative fluorescence data can be represented as a combination of one or more fluorescent dyes or fluorophores, each combination corresponding to a protein marker, cell marker, etc. Quantitative fluorescence data includes compensated data, uncompensated data, and / or combinations of both. Quantitative fluorescence data can be represented as mean fluorescence intensity (MFI). In some embodiments, the cell sorter data may include light scattering data. Light scattering data may include forward light scattering, side light scattering, and / or a combination of both. Light scattering data may be presented as a two-dimensional or three-dimensional linear graph.

[0016] The method may include applying a mapping process to first cell sorter data to determine a representation such as clustering of the first cell sorter data. In some embodiments, the mapping process may include one or more of the following steps: clustering, dimensionality reduction, embedding, nonparametric embedding, or one or more combinations of the above steps. In this document, the term "embedding" refers to a form used to represent a dataset generated by a nonlinear dimensionality reduction algorithm. The term "nonlinear" refers to dimensionality reduction techniques that exclude representations that can be expressed as affine transformations, such as principal component analysis. Examples of nonlinear dimensionality reduction algorithms include: t-distributed random neighborhood embedding (tSNE), uniform manifold approximation and projection (UMAP), pairwise control manifold approximation and projection (PaCMAP), isometric eigenmaps, Isomap, locally linear embedding, and / or related methods. In this document, the term "parametric embedding" refers to embedding or an approximation of embedding, where the coordinates in the dimensionality reduction space are determined by a computational model of individual data points from the dataset. The embedded coordinates may correspond to a point on a scatter plot. Embeddings may include uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other types of nonlinear embeddings, or combinations thereof. The computational model may include a trained machine learning model, such as a trained neural network. In some embodiments, a mapping process can be used to transform data with a high number of dimensions into a representation of transformed data with a low number of dimensions. For example, in some embodiments, the mapping process can transform three-dimensional data into two-dimensional data. In some embodiments, the mapping process is capable of transforming high-dimensional data (including data with 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20 or more dimensions) into two-dimensional or three-dimensional data. In some embodiments, the transformed data includes context-optimized data. In some embodiments, the low-dimensional transformed data representation is optimized to preserve one or more features from the high-dimensional representation. In some embodiments, the data representation includes one or more data clusters, such as cell sorter data for separation into subsequent data representations. In some embodiments, the representation of the first cell sorter data and the representation of the second cell sorter data each include a corresponding two-dimensional chart, a three-dimensional chart, or any combination of the above forms.

[0017] The method may include training a machine learning algorithm based on first cell sorter data and a first representation of the first cell sorter data, for example, using a trained neural network. In some embodiments, the neural network comprises one or more artificial neural networks, convolutional neural networks, recurrent neural networks, or one or more combinations thereof. The method may include training the neural network without pre-determining the first representation of the first cell sorter data, thereby obtaining both a representation of the first cell sorter data and a trained neural network that generates that representation. In some embodiments, the representation includes parameterized embeddings.

[0018] In some embodiments, the method includes receiving second cell sorter data. The cell sorter data may include fluorescence data, light scattering data, or a combination of both. The second cell sorter data may also include microscopy data, hyperspectral imaging data, multispectral data, high-dimensional vector data, or any one or more combinations of the above. In some embodiments, the cell sorter data may include quantitative fluorescence data, presented as the number of antibodies bound per cell, antibody binding capacity (ABC), number of equivalent soluble fluorescent dye molecules (MESF), or one or more other quantitative fluorescence indicators. In some embodiments, the quantitative fluorescence data includes at least one fluorescence signal from one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently conjugated antibodies, or one or more combinations thereof.

[0019] The method may include: determining a representation of second cell sorter data based on a trained machine learning model. In some embodiments, the step of determining the representation of the second cell sorter data includes sorting the second cell sorter data using a field-programmable gate array optimized for performing real-time sorting decisions on the cell sorter. In some embodiments, the step of determining the representation of the second cell sorter data includes processing the machine learning model at a frequency of approximately 100,000 events per second or higher. This processing may be performed in parallel. This processing may not be performed in parallel.

[0020] In some embodiments, the method further includes receiving user input indicating one or more groupings associated with a representation of first cell sorter data, wherein the one or more groupings include one or more signal clustering or gating. In some embodiments, the one or more groupings include one or more signal clustering or gatings with one or more parameters. For example, the grouping method may include clustering or gating operations for removing duplicate samples, screening for viable samples based on specific light scattering values, screening for samples expressing one or more specific lineage markers, or a combination of the above operations. The one or more specific lineage markers may include any phenotypic marker known in the art. For example, the one or more lineage markers may include any known lymphocytes, monocytes, neutrophils, eosinophils, basophils, or one or more combinations thereof. The one or more lineage markers may include any markers applicable to tumor cells, cancer cells, inflammatory cells, etc.

[0021] In some embodiments, the method may further include classifying the data from the second cell sorter based on one or more of the groupings described above and the machine learning model. In some embodiments, the method may also include controlling a device such as a cell sorter to sort a portion of the sample based on the classification results. For example, the device may sort a portion of the cell sample to separate cell subpopulations. These cell subpopulations may express one or more biomarkers of interest, such as one or more biomarkers of phenotypes of interest. For example, the cell subpopulation may express biomarkers of lymphocytes, monocytes, neutrophils, eosinophils, basophils, or a combination thereof. In some embodiments, the cell subpopulation may express any suitable biomarkers of tumor cells, cancer cells, inflammatory cells, etc.

[0022] Another example method may include receiving cell sorter measurement data associated with one or more sensors of a cell sorter device. In some embodiments, the method includes: inputting the received cell sorter data into a machine learning model to determine a representation of the cell sorter data associated with a mapping process. In some embodiments, the cell sorter data representation includes one or more clusters of data for separation into subsequent data representations. In some embodiments, the cell sorter data representation includes a two-dimensional graph, a three-dimensional graph, and / or a combination of both. The representation may include an embedding method, such as parameterized embedding. The method may include an output that causes the cell sorter data representation. In some embodiments, the output-causing step may include one or more of the following operations: displaying the cell sorter data representation, sending the cell sorter data representation to a computing device via a network, prompting an update to a user interface used to control the cell sorter to present measurement results, or a combination of the above operations. The cell sorter data representation may include a scatter plot. In some embodiments, each point on the scatter plot corresponds to a particle or cell. The method may also include receiving user input based on the output representation of the cell sorter data and sorting multiple particles, including multiple cells or a portion of a cell sample, based on the user input. In some embodiments, user input includes setting a gating threshold based on cell sorter data. The gating instruction or gating threshold may include a parameterized embedding aspect of the gating graph drawn by the user based on the first cell sorter data. The user-drawn gating may consist of one or more graphs, such as polygons, ellipses, closed curves, etc.

[0023] Another example method involves determining one or more configuration parameters based on user input for training a machine learning model for an instrument such as a cell sorter. The configuration parameters may include specifications for defining the machine learning model. In some embodiments, the specifications may include weight data. The method may include receiving first cell sorter data, such as cell sorter data from an instrument (e.g., a cell sorter device).

[0024] The method may include: determining an initial representation of the first cell sorter data by applying a mapping process to the first cell sorter data. The data representation may include, for example, a two-dimensional chart, a three-dimensional chart, and / or a combination of both. The mapping process may include one or more of the following steps: clustering, dimensionality reduction, embedding, nonparametric embedding, and / or one or more combinations thereof. Embedding may include uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other types of nonlinear embedding, or combinations thereof. In some embodiments, the mapping process transforms high-dimensional data into a low-dimensional data representation.

[0025] The method may include training based on first cell sorter data and some representation of the first cell sorter data, for example, using a machine learning model constructed using a trained neural network. The machine learning model may include a deep learning model. For example, the machine learning model may include a neural network. In some embodiments, the neural network may include one or more artificial neural networks, convolutional neural networks, recurrent neural networks, or one or more combinations thereof.

[0026] The method may include prompting the storage of a machine learning model. The machine learning model may be stored by a computer processor communicating with a cell sorter. The machine learning model may be stored using short-term storage components. The machine learning model may be stored using long-term storage components. In some embodiments, the machine learning model may be stored using one or more components, such as a field-programmable gate array (FPGA). The computer processor may be configured to analyze second cell sorter data using the stored machine learning model, including cell sorter data received from the instrument. The second cell sorter data may include microscopic data, hyperspectral imaging data, multispectral data, high-dimensional vector data, or one or more combinations thereof. The cell sorter data may include quantitative fluorescence data, representing the number of antibodies bound to each cell or antibody binding capacity (ABC), equivalent soluble fluorescent dye molecules (MESF), one or more other quantitative fluorescence indicators, or one or more combinations thereof. The cell sorter data may include one or more fluorescence signals from one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, or one or more combinations thereof.

[0027] In some aspects, this disclosure provides a method for classifying samples containing multiple particles, the method comprising: determining a representation of first cell sorter data based on applying a trained machine learning model to first cell sorter data. The representation may include an embedding method, such as parametric embedding. Parametric embedding may include a two-dimensional scatter plot, where each point corresponds to a cell or particle. In some embodiments, the method may include the step of: training a machine learning model based on one or more features in the first cell sorter data, user input, a first representation of the first cell sorter data, or a combination thereof. In some embodiments, the method may include the step of receiving second cell sorter data associated with a second portion of one or more cell samples. In some embodiments, the method may include a step of sorting a portion of the sample based on a classification result derived from the trained machine learning model and / or a user-provided gating instruction.

[0028] In some aspects, the present invention discloses a method for sorting samples containing multiple particles or cells. In some embodiments, the method may include collecting first cell sorter data associated with a first portion of one or more cell samples. In some embodiments, the method may include determining a first representation of the first cell sorter data based on applying a mapping process to the first cell sorter data. In some embodiments, the method may include training a machine learning model based on one or more first cell sorter data, user input, a first representation of the first cell sorter data, or any combination thereof. In some embodiments, the method may include determining a parameterized embedding of the first cell sorter data based on first measurement data and the trained machine model. Embodiments of the method may include receiving one or more gating instructions from a user using the first cell sorter data, the parameterized embedding of the first cell sorter data, or a combination thereof. Embodiments of the method may include receiving second cell sorter data associated with a second portion of one or more cell samples. Embodiments of the method may include determining the embedding coordinates of the second cell sorter data using a trained machine learning model. Embodiments of the method may include classifying the second cell sorter data based on the embedding coordinates of the second cell sorter data and gating instructions provided by the user. Implementations of the method may include: sorting a portion of the samples based on classification criteria.

[0029] In some respects, the device covered by this disclosure includes one or more processors; and a memory storing instructions that, when executed by one or more processors, cause the device to perform any of the methods disclosed herein.

[0030] In some respects, this disclosure provides a non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause the device to perform any of the methods disclosed herein.

[0031] In some aspects, this disclosure provides a system including a cell sorter comprising one or more sensors configured to generate cell sorter data; and a computing device comprising one or more processors and a memory storing instructions which, when executed by the one or more processors, cause the computing device to perform any of the methods disclosed herein. For example, the memory is capable of storing one or more weights associated with one or more trained machine learning models, including, for example, one or more trained neural networks as described herein. In this document, the term “weight” in relation to a neural network refers to all parameter values ​​and network structure definitions necessary for processing input data to obtain output values ​​through the neural network.

[0032] The cell sorting system disclosed in this invention exhibits superior performance compared to conventional methods. For example, existing cell sorting strategies are based on clusters derived from nonlinear embeddings of cell sorting data, relying on gating algorithms that plot a series of two-dimensional gatings based on the expression levels of cell markers. Here, the term "gating localization" refers to the process of locating a series of two-dimensional gatings based on cell marker expression levels, scatter data, fluorescence measurements, and / or combinations thereof (which approximate the gating shape on the embedding). These algorithms are not precise enough to utilize the relationships between three or more cell markers within a single gating region. This invention addresses this problem by using a machine learning model (e.g., a neural network) to learn a parameterized representation of the embedding. This model, combined with an integrated, efficient computing architecture, enables real-time embedding of new data points and sorting based on the embedding coordinates. The related embodiments disclosed in this invention thus improve cell sorting technology, for example, by improving the computer technology supporting such cell sorting instruments.

[0033] The various methods and cell sorting systems disclosed in this invention can utilize machine learning models (e.g., trained neural networks) to improve upon conventional methods, thereby achieving higher sorting efficiency, more accurate algorithms, and faster cell sorting results. This model can process and sort cell sorting data in real time and display the data graphically. These technical advantages are unattainable by conventional methods, and all system users, including those described, can benefit from them, for example, by assisting users in performing technical tasks, such as real-time high-throughput cell sorting, through guided human-computer interaction. The technical features of the various embodiments disclosed in this invention are clearly groundbreaking in the field of cell sorting instruments, as are the combinations of features of the various embodiments disclosed in this invention. Therefore, this disclosure introduces functionality that is impossible to perform with conventional computing devices and humans. As used herein, the term "weight," when used in relation to neural networks, refers to all parameter values ​​and network structure definitions necessary for processing input data to obtain output values ​​through the neural network.

[0034] The embodiments disclosed in this invention can improve cell sorting / cell sorter technology, for example, by improving computer technology that supports the cell sorter.

[0035] In the following detailed description, reference is made to the accompanying drawings, which form part of the detailed description, wherein like reference numerals always indicate like parts, and practical embodiments are illustrated in the drawings by way of illustration. It should be understood that other embodiments may be utilized, and structural or logical element changes may be made without departing from the scope of this disclosure. Therefore, the following detailed description should not be regarded as limiting.

[0036] Various operations can be described as multiple discrete actions or operations in a manner most conducive to understanding the subject matter disclosed herein. However, the order of description should not be construed as implying that these operations necessarily depend on a specific order. Specifically, these operations may be performed in an order different from the order in which they are presented. The described operations may be performed in an order different from the described embodiments. Various additional operations may be performed, and / or the described operations may be omitted in additional embodiments.

[0037] For the purposes of this disclosure, the phrases "A and / or B" and "A or B" mean (A), (B), or (A and B). For the purposes of this disclosure, the phrases "A, B and / or C" and "A, B or C" mean (A), (B), (C), (A and B), (A and C), (B and C), or (A, B and C). Although some elements may be mentioned in the singular (e.g., "a processing device"), any suitable element may be represented by multiple instances of that element, and vice versa. For example, a set of operations described as being performed by a processing device can be implemented using different operations performed by different processing devices.

[0038] This specification uses the phrases "embodiments," "various embodiments," and "some embodiments," each referring to one or more embodiments of the same or different embodiments. Furthermore, the terms "comprising," "including," "having," etc., used with respect to embodiments of this disclosure are synonymous. When used to describe a range of dimensions, the phrase "between X and Y" indicates a range encompassing both X and Y. As used herein, "apparatus" can refer to any single device, a collection of devices, a portion of a device, or a collection of portions of devices. The drawings are not necessarily drawn to scale.

[0039] Figure 1A This is a block diagram of a cell sorter support module 1100 for performing supporting operations based on various embodiments. The cell sorter support module 1100 can be implemented by a circuit system (e.g., including electrical and / or optical components) such as a programmed computing device. The logic elements of the cell sorter support module 1100 can be included in a single computing device or distributed across multiple computing devices that need to communicate with each other. References herein Figure 4 The computing device 4000 discusses examples of computing devices that can be implemented individually or in combination to support the cell sorter module 1100, and provides a reference.Figure 5 The Cell Sorter Support System 5000 discusses an example of an interconnected computing device system, in which the Cell Sorter Support Module 1100 can be implemented across one or more computing devices.

[0040] The cell sorter support module 1100 may include a first receiving logic element 1102, a first determining logic element 1104, a training logic element 1106, a second receiving logic element 1108, and a second determining logic element 1110. A specific implementation of the cell sorter support module 1100 may also include a third receiving logic element 1112, a third determining logic element 1014, and a triggering logic element 1116. As used herein, the term "logic element" can include means for performing a set of operations associated with said logic element. For example, any of the logic elements included in the support module 1100 may be implemented by one or more computing devices programmed with instructions to cause one or more processing devices of the computing device to perform an associated set of operations. One or more computing devices may include, for example, one or more field-programmable gate arrays (FPGAs). In one particular embodiment, a logic element may include one or more non-transitory computer-readable media having instructions thereon, wherein when executed by one or more processing devices of the one or more computing devices, said instructions cause the one or more computing devices to perform an associated set of operations. As used herein, the term "module" can refer to a collection of one or more logical units that collectively perform functions associated with said module. Different logic elements within a module may take the same form or may take different forms. For example, some logic elements in a module may be implemented by a general-purpose programmable processing device, while others may be implemented by an application-specific integrated circuit (ASIC). In another example, different logic elements in a module may be associated with different sets of instructions executed by one or more processing devices. A module may not include all the logic units depicted in the associated figures; for example, a module may include a subset of the logic units depicted in the associated figures when it performs a subset of the operations discussed herein with reference to the module.

[0041] The first receiving logic element 1102 can receive first cell sorting data from the cell sorting device. The data acquired by the cell sorting device can be presented in the form of fluorescence signal data, light scattering data, and / or a combination of both, allowing fluorescence signals to be obtained from one or more cells in the sample. Alternatively, the first cell sorting device data can include fluorescence signal data, such as microscopic data, hyperspectral imaging data, multispectral imaging data, high-dimensional vector data, or one or more combinations thereof. The cell sorting device data may include quantitative fluorescence data, representing one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations thereof.

[0042] The first determining logic element 1104 can determine a first representation of the first cell sorter data by applying a mapping process to the first cell sorter data. The mapping process may include one or more of the following steps: clustering, dimensionality reduction, embedding, nonparametric embedding, or one or more combinations thereof. Embedding may include one or more methods or combinations thereof, such as Uniform Manifold Approximation and Projection (UMAP), t-Distributed Random Neighborhood Embedding (t-SNE), or other types of nonlinear embedding. The representation may contain one or more data clusters for separation into subsequent data representations. The data representation may be a two-dimensional graph, a three-dimensional graph, and / or a combination of both.

[0043] Training logic element 1106 can train a machine learning model based on first cell sorter data and a first representation of the first cell sorter data. The machine learning model may include a learning mechanism for a trained neural network, which is a mapping process for generating one or more representations of data for a cell sorter measurement session. The neural network can be trained without pre-determining the first representation of the first cell sorter data, obtaining both the representation of the first cell sorter data and the trained neural network that generates that representation. The neural network may include one or more of artificial neural networks, convolutional neural networks, recurrent neural networks, or combinations thereof.

[0044] The second receiving logic element 1108 can receive second cell sorting data from the cell sorting device. The second cell sorting data may include cell sorting data received in the form of fluorescence signal data, light scattering data, and / or a combination of both. Alternatively, the second cell sorting data may include fluorescence signal data, such as microscopic data, hyperspectral imaging data, multispectral imaging data, high-dimensional vector data, or any combination of the above. Fluorescence signals can be obtained from one or more cells in the sample. The fluorescence signal data may include quantitative fluorescence data, which represents the number of antibodies bound to each cell or antibody binding capacity (ABC), equivalent soluble fluorescent dye molecules (MESF), one or more other quantitative fluorescence indicators, or one or more combinations of the above.

[0045] The second determining logic element 1110 may include determining a representation of the second cell sorter data based on a trained machine learning model. The representation may include one or more clusters of data for separation into subsequent data representations. The data representation may be a two-dimensional chart, a three-dimensional chart, and / or a combination of both. A two-dimensional or three-dimensional chart may include a scatter plot. Each point on the scatter plot corresponds to a particle or cell. Determining the representation of the second cell sorter data may include sorting the second cell sorter data using a field-programmable gate array (FPGA). The FPGA may be optimized for real-time sorting decisions on the cell sorter. The sorting decision can be made in approximately 100 microseconds, or between approximately 100 microseconds and approximately 300 microseconds, or between approximately 300 microseconds and approximately 500 microseconds, or between approximately 500 microseconds and approximately 700 microseconds, or between approximately 700 microseconds and approximately 900 microseconds, or between approximately 900 microseconds and approximately 1 millisecond, or between approximately 1 millisecond and approximately 2 milliseconds, or between approximately 2 milliseconds and approximately 5 milliseconds, or between approximately 5 milliseconds and approximately 10 milliseconds, or between approximately 10 milliseconds and approximately 25 milliseconds, or between approximately 25 milliseconds and approximately 50 milliseconds, or between approximately 50 milliseconds and approximately 75 milliseconds, or between approximately 75 milliseconds and approximately 100 milliseconds, or between approximately 100 milliseconds and approximately 500 milliseconds, or between approximately 500 milliseconds and approximately 1 second, or between approximately 1 second and approximately 2 seconds, or between approximately 2 seconds and approximately 5 seconds, or between approximately 5 seconds and approximately 10 seconds, including any and all increments within the above ranges.

[0046] The third receiving logic element 1112 can receive user input indicating one or more packets associated with the representation of the first cell sorter data.

[0047] The third determining logic element 1114 can determine the classification of the second cell sorter data based on one or more sets of data and a machine learning model.

[0048] The causal logic element 1116 can cause the device to sort a portion of the sample based on a classification determined by the third determining logic element 1114. The sample may contain cell samples. For example, the sample may contain samples of one or more cell types, such as lymphocytes, monocytes, neutrophils, eosinophils, basophils, or one or more combinations thereof. In some embodiments, the cell sample may contain samples of tumor cells, cancer cells, inflammatory cells, epithelial cells, endothelial cells, germ cells, somatic cells, etc., or one or more combinations thereof.

[0049] Figure 1B This is a block diagram of a cell sorter support module 1200 for performing supporting operations based on various embodiments. The cell sorter support module 1200 can be implemented by a circuit system (e.g., including electrical and / or optical components) such as a programmed computing device. The logic elements of the cell sorter support module 1200 can be included in a single computing device or distributed across multiple computing devices that need to communicate with each other. References herein Figure 4 The computing device 4000 discusses examples of computing devices that can implement the cell sorter support module 1200 alone or in combination, and this document references Figure 5 The support system 5000 discusses examples of cell sorter support systems in which the cell sorter support module 1100 can be implemented across one or more computing devices.

[0050] Cell sorter support module 1200 may include a first receiving logic element 1202, a first determining logic element 1204, and a causal logic element 1206. In some embodiments, cell sorter support module 1200 may optionally include a second receiving logic element 1208 (not shown). The term "logic element" as used herein can include means for performing a set of operations associated with said logic element. For example, any of the logic elements included in support module 1100 may be implemented by one or more computing devices programmed with instructions to cause one or more processing devices of the computing device to perform an associated set of operations. In one particular embodiment, a logic element may include one or more non-transitory computer-readable media having instructions thereon, wherein, when executed by one or more processing devices of the one or more computing devices, the instructions cause the one or more computing devices to perform an associated set of operations. The term "module" as used herein can refer to a collection of one or more logical units that collectively perform functions associated with said module. Different logic elements in a module may take the same form or different forms. For example, some logic elements in a module may be implemented by a general-purpose programming processing device, while other logic elements in the module may be implemented by an application-specific integrated circuit (ASIC). In another example, different logical elements in a module may be associated with different sets of instructions executed by one or more processing devices. A module may not include all the logical units depicted in the associated figures; for example, a module may include a subset of the logical units depicted in the associated figures when it performs a subset of operations discussed herein with reference to the module.

[0051] The first receiving logic element 1202 can receive cell sorting measurement data associated with one or more sensors of the cell sorting device. The instrumentation may include cell sorting equipment. The sensors may include one or more fluorescence detectors. Cell sorting data may include cell sorting data received in the form of fluorescence signal data. For example, fluorescence signal data may include hyperspectral imaging data, high-dimensional vector data, or one or more combinations thereof. Fluorescence signals can be acquired from one or more cells in a sample. The one or more cells may be modified to express one or more fluorescent proteins. One or more of the cells may be labeled with one or more fluorescent markers or other fluorescent markers. Fluorescence signal data may include quantitative fluorescence data, representing one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations thereof.

[0052] The first determining logic element 1204 can determine the representation of the cell sorter data based on the operation of inputting cell sorter data into a machine learning model. The representation can include one or more clusters of data for separation into subsequent data representations. The data representation can be a two-dimensional graph, a three-dimensional graph, and / or a combination of both. A two-dimensional or three-dimensional graph can include a scatter plot. Each point on the scatter plot corresponds to a particle or cell. The machine learning model can include a learning mechanism of a trained neural network that is a mapping process for generating one or more representations of the data for a cell sorter measurement session. The mapping process can include one or more of the following steps: clustering, dimensionality reduction, parameterized embedding, non-parameterized embedding, and / or any combination of the above steps. Embedding can include uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other types of nonlinear embedding, or combinations thereof. The neural network can include one or more of artificial neural networks, convolutional neural networks, recurrent neural networks, or one or more combinations thereof.

[0053] The causal logic element 1206 can cause the device to classify a portion of a sample based on a classification determined by the third determining logic element 1014. The sample may contain cell samples. For example, the sample may contain samples of one or more cell types, such as lymphocytes, monocytes, neutrophils, eosinophils, basophils, or one or more combinations thereof. In some embodiments, the cell sample may contain tumor cells, cancer cells, inflammatory cells, etc., which may contain any one or more combinations of the aforementioned cell types.

[0054] The second receiving logic element 1208 can receive second cell sorting data from the cell sorting device. The second cell sorting data may include received cell sorting data in the form of fluorescence signal data, light scattering data, and / or a combination of both. Fluorescent particles can be obtained from one or more cells in the sample. Fluorescent signals can be obtained from one or more cells in the sample. The cells may contain cells modified to express one or more fluorescent proteins. The cells may contain cells labeled with one or more fluorescent agents or fluorescent probes. The fluorescence signal data may include quantitative fluorescence data, representing one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations of the above.

[0055] Figure 1CThis is a block diagram of a cell sorter support module 1300 for performing supporting operations based on various embodiments. The cell sorting support module 1300 can be implemented by a circuit system (e.g., including electrical and / or optical components) such as a programmed computing device. The logic elements of the cell sorter support module 1300 can be included in a single computing device or distributed across multiple communicating computing devices as needed. References herein Figure 4 The computing device 4000 discusses examples of computing devices that can implement the cell sorter support module 1300 alone or in combination, and this document references Figure 5 The Cell Sorting Support System 5000 discusses system examples of interconnected computing devices in which the Cell Sorter Support Module 1300 can be implemented across computing devices.

[0056] The cell sorter support module 1300 may include a first determining logic element 1302, a first receiving logic element 1304, a second determining logic element 1306, a training logic element 1308, and a causal logic element 1310. As used herein, the term "logic element" can include means for performing a set of operations associated with said logic element. For example, any logic element included in the support module 1300 may be implemented by one or more programmable computing devices that issue instructions to one or more processing devices of the computing device to perform an associated set of operations. The one or more computing devices may include, for example, one or more field-programmable gate arrays (FPGAs). In one particular embodiment, a logic element may include one or more non-transitory computer-readable media having instructions thereon, which, when executed by one or more processing devices of the one or more computing devices, cause the one or more computing devices to perform an associated set of operations. As used herein, the term "module" can refer to a collection of one or more logical units that collectively perform functions associated with said module. Different logic elements in a module may take the same form or different forms. For example, some logic elements in a module may be implemented by a general-purpose programming processing device, while other logic elements in the module may be implemented by an application-specific integrated circuit (ASIC). In another example, different logical elements in a module may be associated with different sets of instructions executed by one or more processing devices. A module may not include all the logical units depicted in the associated figures; for example, a module may include a subset of the logical units depicted in the associated figures when it performs a subset of operations discussed herein with reference to the module.

[0057] The first determining logic element 1302 can determine one or more configuration parameters for training machine learning equipment based on user input. The configuration parameters may include one or more parameters specifying the structure of the machine learning model. For example, one or more parameters may include information specifying the depth of the neural network, the size of the neural network / layer, activation functions, etc. The depth of the neural network may include 1 hidden layer, 2 hidden layers, 3 hidden layers, 4 hidden layers, 5 hidden layers, 6 hidden layers, etc. The neural network may contain up to 10 hidden layers, up to 100 hidden layers, and so on.

[0058] Each layer can have 1 node, 2 nodes, 3 nodes, and so on. A neural network can contain a maximum of 10 nodes per layer, a maximum of 100 nodes per layer, and so on. The number of nodes in each layer may or may not be the same. The activation function can be selected from one or more activation functions, such as the binary step function, linear activation function, sigmoid / logistic activation function, the derivative of the sigmoid activation function, the Tanh function (hyperbolic tangent), the Tanh gradient activation function, the Rectified Unit (ReLU) activation function, the hard sigmoid function, the leaky ReLU function, other nonlinear activation functions, or one or more combinations of the above.

[0059] The configuration parameters may include one or more training and / or optimization parameters. Training and / or optimization parameters may include one or more specific distance metrics, loss functions, optimization algorithms and their hyperparameters, learning rates, regularization terms, instructions on the use of the lost layer, early stopping decision parameters, and accuracy thresholds, or one or more combinations of the above (including one or more other metrics that the trained network deems usable).

[0060] The configuration parameters may include at least one or more preprocessing instructions for the data used by the machine learning algorithm. These preprocessing instructions may include data selection methods such as gating thresholds, fluorescence intensity cutoff values, selection of specific measurements to be used as input to the machine learning algorithm, and / or one or more combinations thereof. The preprocessing instructions may also include applying one or more scaling functions and associated parameters. These scaling functions may include one or more double exponential, logarithmic, logarithmic, and / or hyperlogarithmic transformations.

[0061] The configuration parameters may include selecting the computing hardware used to perform the computational work required to train machine learning algorithms. The computing hardware may include, but is not limited to, a central processing unit (CPU), a graphics processing unit (GPU), or other industry-recognized applicable computing systems.

[0062] The first receiving logic element 1304 can receive first cell sorting data from the cell sorting device. The data acquired by the cell sorting device can be presented in the form of fluorescence signal data, light scattering data, and / or a combination of both, from one or more cells in the sample. The cell sorting device data may include quantitative fluorescence data, which represents one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations of the above.

[0063] The second determining logic element 1306 can determine a first representation of the first cell sorter data by applying a mapping process to the first cell sorter data. The mapping process may include one or more of the following steps: clustering, dimensionality reduction, embedding, nonparametric embedding, or one or more combinations thereof. Embedding may include uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other types of nonlinear embedding, or combinations thereof. The representation may contain one or more data clusters for separation into subsequent data representations. The data representation may be a two-dimensional graph, a three-dimensional graph, and / or a combination of both.

[0064] Training logic element 1308 can train a machine learning model based on first cell sorter data and its initial representation. This machine learning model may include a learning mechanism for a trained neural network, which is a mapping process for generating one or more representations of data used in a cell sorter measurement session. The neural network can be trained without pre-determining the initial representation of the first cell sorter data, obtaining both the representation of the first cell sorter data and the trained neural network that generates that representation. The neural network may include one or more artificial neural networks, convolutional neural networks, recurrent neural networks, or a combination of these.

[0065] The causal logic element 1310 may cause the storage of a machine learning model, wherein a computer processor (such as an FPGA) communicating with the instrument is configured to use the stored machine learning model to analyze second cell sorter data received from the instrument.

[0066] Figure 2A This is a flowchart of a method 2100 for performing supporting operations based on various embodiments. Although the operation of method 2100 can be described with reference to specific embodiments disclosed herein (e.g., references are made herein to…),… Figure 1A , 1B The cell sorter support module 1000 discussed in 1C, and referenced in this article... Figure 3 The graphical user interface 3000 discussed herein is referenced in this document.Figure 4 The computing device discussed is 4000, and the references in this article are also included. Figure 5 The cell sorting support system 5000 is discussed, but the method 2100 can be used in any suitable environment to perform any suitable support operation. In Figure 2, each operation is shown once in a specific order, but these operations can be reordered and / or repeated as needed. For example, different operations can be performed in parallel as appropriate.

[0067] At 2102, a first operation can be performed. For example, a first logic element 1102 of support module 1100 can perform the operation at 2102. The first operation may include receiving first cell sorter data.

[0068] At 2104, a second operation can be performed. For example, a second logic element 1104 of the support module 1100 can perform the operation at 2104. The second operation may include determining a representation of the first cell sorter data.

[0069] At 2106, a third operation can be performed. For example, the third logic element 1106 of the support module 1100 can perform the operation at 2106. The third operation may include training a machine learning model based on the first cell sorter data and a first representation of the first cell sorter data. The machine learning model may include a neural network. The neural network can be trained without pre-determining the representation of the first cell sorter data, thereby obtaining both the representation of the first cell sorter data and a trained neural network capable of generating that representation.

[0070] At 2108, a fourth operation can be performed. For example, the fourth logic element 1108 of the support module 1100 can perform the operation at 2108. The third operation may include receiving data from the second cell sorter.

[0071] At 2110, a fifth operation can be performed. For example, the fifth logic element 1110 of the support module 1100 can perform the operation at 2110. The fifth operation may include determining a representation of the second cell sorter data based on a trained machine learning model.

[0072] At 2112, a sixth operation can be performed. For example, the sixth logic element 1112 of support module 1100 can perform the operation of 2112. The sixth operation may include receiving user input indicating one or more groups associated with a representation of the second cell sorter data.

[0073] At 2114, a sixth operation can be performed. For example, the seventh logic element 1114 of the support module 1100 can perform the operation at 2114. The sixth operation may include classifying and determining the data from the second cell sorter.

[0074] At 2116, an eighth operation can be performed. For example, the eighth logic element 1116 of the support module 1100 can perform the operation at 2116. The eighth operation may include causing the device to sort a portion of the sample.

[0075] Figure 2B This is a flowchart of the execution support operation method 2200 based on various embodiments. Although the operation of method 2200 can be described with reference to specific embodiments disclosed herein (e.g., referred to herein...), Figure 1A , 1B The cell sorter support module 1100-1300 discussed in 1C, and referenced in this article... Figure 3 The graphical user interface 3000 discussed herein is referenced in this document. Figure 4 The computing device discussed is 4000, and / or referenced herein. Figure 5 The cell sorter support system discussed is 5000), but the method 2200 can be used in any suitable environment to perform any suitable support operation. Figure 2B In this context, each operation is shown once in a specific order, but these operations can be reordered and / or repeated as needed and as appropriate. For example, different operations can be performed in parallel as appropriate.

[0076] At 2202, a first operation can be performed. For example, a first logic element 1202 of support module 1200 can perform the operation at 2202. The first operation may include receiving cell sorter data associated with one or more sensors of the instrumentation.

[0077] At 2204, a second operation can be performed. For example, a second logic element 1204 of the support module 1200 can perform the operation at 2204. The second operation may include determining a representation of the cell sorter data associated with the mapping process. This representation can be generated by a machine learning model without prior knowledge of the information in the first representation data. This representation can be generated by a trained machine learning model.

[0078] At 2206, a third operation can be performed. For example, the third logic element 1206 of the support module 1200 can perform the operation at 2206. The third operation may include prompting the display of a representation of cell sorter data.

[0079] At 2208, a fourth operation can be performed. For example, the fourth logic element 1208 of the support module 1200 can perform the operation at 2208. The fourth operation may include receiving user input based on the output representation of cell sorter data, and sorting multiple particles, including, for example, multiple cells or a portion of a cell sample, based on the user input.

[0080] Figure 2C Flowcharts of a method 2300 for performing supporting operations based on various embodiments. Although the operation of method 2300 can be described with reference to specific embodiments disclosed herein (e.g., references herein to...), Figure 1A , 1B The cell sorter support module 1100-1300 discussed in 1C, and referenced in this article... Figure 3 The graphical user interface 3000 discussed herein is referenced in this document. Figure 4 The computing device discussed is 4000, and / or referenced herein. Figure 5 The cell sorter support system discussed is 5000), but the method 2300 can be used in any suitable environment to perform any suitable support operation. Operation in... Figure 2C The operations are described once and in a specific order, but may be reordered and / or repeated as needed and as appropriate (e.g., different operations may be performed in parallel as appropriate).

[0081] At 2302, a first operation can be performed. For example, a first logic element 1302 of support module 1300 can perform the operation at 2302. The first operation may include determining one or more configuration parameters for machine learning of the instrument based on user input.

[0082] At 2304, a second operation can be performed. For example, the second logic element 1304 of the support module 1300 can perform the operation at 2304. The second operation may include receiving first cell sorter data from the instrumentation.

[0083] At 2306, a third operation can be performed. For example, the third logic element 1306 of the support module 1300 can perform the operation at 2306. The third operation may include: applying a mapping process to the first cell sorter data to determine a representation such as clustering of the first cell sorter data.

[0084] At 2308, a fourth operation can be performed. For example, the fourth logic element 1308 of support module 1300 can perform the operation at 2308. The fourth operation may include training a machine learning model based on the first cell sorter data and a first representation of the first cell sorter data. The fourth operation may include training a machine learning model that determines the representation of the first cell sorter data without prior determination of the representation of the first cell sorter data. The machine learning model may include a neural network. The neural network can be trained without prior determination of the representation of the first cell sorter data, obtaining both the representation of the first cell sorter data and the trained neural network that generates that representation. The representation may include an embedding method.

[0085] At 2310, a fifth operation can be performed. For example, the fifth logic element 1310 of support module 1300 can perform the operation at 2310. The fifth operation may include storing a machine learning model using a computer processor communicating with the instrument.

[0086] The cell sorting instrument support methods disclosed in this article may include interaction with human users (e.g., through the methods mentioned herein). Figure 5 The user's local computing device 5020 is implemented in this context. The interaction may include providing the user with information (e.g., information about...). Figure 5 This includes operational information about cell sorters, including the cell sorter 5010; information about the sample being analyzed or other tests or measurements performed by the cell sorter; information retrieved from local or remote databases, etc.; or options for providing users with input commands (e.g., controls including...). Figure 5 The operation of cell sorters, including the cell sorter 5010, or the control of data analysis generated by the cell sorter, querying (e.g., querying a local or remote database), etc., are permitted. In some embodiments, these interactions may be performed via a graphical user interface (GUI) including a display device (e.g., referenced herein). Figure 4 A visual display on a display device 4010 (discussed herein) that provides output to a user and / or prompts the user to provide input (e.g., via reference herein). Figure 4 Other I / O devices discussed 4012 include one or more input devices, such as a keyboard, mouse, touchpad, or touchscreen. The cell sorter support system disclosed herein may include any suitable GUI for interacting with a user. In some embodiments, the visual display interface contains one or more representations of cell sorter data. The representation may include at least one two-dimensional or three-dimensional graph. In some embodiments, the one or more representations are generated in real time.

[0087] Figure 3 A GUI 3000 based on various embodiments is described, which can be used to perform some or all of the supporting methods disclosed herein. As described above, the GUI 3000 can be set in a cell sorting instrument support system (e.g., as referenced herein). Figure 5 The computing device (e.g., the cell sorting instrument support system 5000 discussed herein) refers to the computing device (e.g., the one referenced in this article). Figure 4 The display device (e.g., the computing device 4000 discussed herein) is a display device (referring to the present document). Figure 4 The display device discussed is 4010, and the user can use any suitable input device (e.g., included in the references herein). Figure 4The other I / O devices discussed (any of the input devices in the 4012 discussion) and input technologies (e.g., cursor movement, motion capture, facial recognition, gesture detection, voice recognition, button actuation, etc.) are used to interact with the GUI 3000.

[0088] The GUI3000 may include a data display area 3002, a data analysis area 3004, a cell sorter control area 3006, and a setting area 3008. Figure 3 The specific number and arrangement of the areas depicted are merely illustrative, and any number and arrangement of areas including any desired features may be included in GUI3000.

[0089] Data display area 3002 can display data generated by the cell sorter, for example, including data included here via reference. Figure 5 The data generated by the cell sorter 5010 is discussed. For example, data display area 3002 can display the range of fluorescence intensity, gating information, clustering information, etc.

[0090] Data analysis area 3004 can display the results of data analysis, such as the results of analyzing the data displayed in data display area 3002 and / or other data. For example, data analysis area 3004 can display one or more representations of cell sorter data. In some embodiments, data display area 3002 and data analysis area 3004 can be combined in GUI 3000 (e.g., including data output from the cell sorter and some analysis of the data in a common graphic or area).

[0091] The cell sorter control area 3006 may include options that allow the user to control the cell sorter, as described in this article. Figure 5 The cell sorter 5010 under discussion. For example, the cell sorting control area 3006 may include one or more user interfaces for receiving user input. User input may include setting a gating threshold based on cell sorter data. The set gating threshold may include gating instructions provided by the user, such as drawing a gating based on parameterized embeddings based on cell sorter data. User input may include assigning fluorophores to specific channels and assigning cell markers to fluorophores. User input may also include control parameters related to the instrument's fluid systems, electronic systems, mechanical systems, etc. Control parameters may include values ​​for one or more parameters specified by the user, such as event rate, PMT voltage, etc.

[0092] User input may also include specifying one or more parameters associated with one or more physical devices and / or modes used for sorting cell samples. For example, user input may include specifying multiple receiving containers for collecting the sorted and distinct cell populations. The one or more receiving containers may include one or more tubes with a specific capacity, one or more plates with a specific number of pores, and similar components. User input may also include specifying the number of cells received or to be received in each receiving container. User input may also include specifying the direction of waste liquid.

[0093] User input may include gating instructions, including specifying one or more gating thresholds. These gating instructions may be determined based on one or more of the following data: fluorescence data, light scattering data, fluorescent label expression level data, data output from a machine learning model, embedding output from a machine learning model, parameterized embeddings of cell sorting data generated using a machine learning model, or a combination of the above. One or more gating thresholds or gating mechanisms may be used as a preprocessing step for the machine learning model. In some embodiments, one or more conventional gating schemes may be combined with the gating mechanism used in the machine learning model. For example, a first gating may be specified and applied to distinguish between individual cells and cell clusters, and another gating may be specified and applied based on the machine learning representation of the data. One or more sorting decision paths applied to the user input may include information specifying the use of one or more conventional data representations, one or more machine learning-based data representations, and / or information from both conventional and machine learning-based data representations. Figure 7A , 7B The 7C demonstrates an example of a conventional user-drawn gating scheme based on the expression levels of fluorescent markers in first-cell sorter data.

[0094] Setting up area 3008 may include options that allow a user to control the features and functions of GUI 3000 (and / or other GUIs), and / or perform common computational operations relative to data display area 3002 and data analysis area 3004 (e.g., storing data on a storage device such as referenced herein). Figure 4The discussed storage device 4004 is used for sending data to other users, tagging data, etc. For example, settings area 3008 may include one or more options for setting one or more display parameters. These display parameters may include setting one or more scaling functions for data visualization, such as setting upper or lower limits for the axes of one or more displayed charts. The one or more display parameters may also include setting one or more data storage format parameters. The display parameters may also include setting one or more parameters for exporting data, saving data, exporting one or more models, etc. The computer operation can be performed using an FPGA, etc.

[0095] As described above, the cell sorter support module 1000 can be implemented by one or more computing devices. Figure 4 This is a block diagram of an exemplary computing device 4000 that can be used to perform some or all of the supported methods disclosed herein, based on various embodiments. In some embodiments, the cell sorter support module 1000 may be implemented by a single computing device 4000 or by multiple computing devices 4000. Furthermore, as described below, the computing device 4000 (or multiple computing devices 4000) implementing the cell sorter support module 1000 may be... Figure 5 The device may be a part of one or more of the following: a cell sorter 5010, a user local computing device 5020, a service local computing device 5030, or a remote computing device 5040. The computing device may include an FPGA, etc.

[0096] Figure 4 The computing device 4000 is shown as having multiple components, but any or more of these components may be omitted or repeated based on application and setup requirements. In some embodiments, some or all of the components included in the computing device 4000 may be attached to one or more motherboards and encapsulated in a housing (e.g., including plastic, metal, and / or other materials). In some embodiments, some of these components may be fabricated onto a single system-on-a-chip (SoC) (e.g., the SoC may include one or more processing devices 4002 and one or more storage devices 4004). Additionally, in various embodiments, the computing device 4000 may not include... Figure 4One or more of the components shown may be included, but may include interface circuitry (not shown) for coupling to one or more components using any suitable interface, such as a Universal Serial Bus (USB) interface, a High Definition Multimedia Interface (HDMI) interface, a Controller Area Network (CAN) interface, a Serial Peripheral Interface (SPI) interface, an Ethernet interface, a wireless interface, or any other suitable interface. For example, computing device 4000 may not include display device 4010, but may include display device interface circuitry (e.g., connector and driver circuitry) to which display device 4010 may be coupled.

[0097] The computing device 4000 may include a processing device 4002. The processing device may include one or more processing devices. As used herein, the term "processing device" can refer to any device or part of a device that processes electronic data from registers and / or memory to convert that electronic data into other electronic data that can be stored in registers and / or memory. Processing device 4002 may include one or more digital signal processors (DSPs), field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), central processing units (CPUs), graphics processing units (GPUs), cryptographic processors (dedicated processors that execute cryptographic algorithms within hardware), server processors, or any other suitable processing device.

[0098] The computing device 4000 may include a storage device 4004. The storage device may include one or more storage devices. Storage device 4004 may include one or more memory devices, such as random access memory (RAM) (e.g., static RAM (SRAM) devices, magnetic RAM (MRAM) devices, dynamic RAM (DRAM) devices, resistive RAM (RRAM) devices, or conductive bridged RAM (CBRAM) devices), hard disk drive-based memory devices, solid-state memory devices, network drives, cloud drives, or any combination of memory devices. In some embodiments, storage device 4004 may include memory sharing a die with processing device 4002. In such embodiments, the memory may be used as cache memory and may include, for example, embedded dynamic random access memory (eDRAM) or spin-transfer torque magnetic random access memory (STT-MRAM). In some embodiments, storage device 4004 may include a non-transitory computer-readable medium having instructions thereon that, when executed by one or more processing devices (e.g., processing device 4002), cause computing device 4000 to perform any suitable method or portion thereof of the methods disclosed herein.

[0099] The computing device 4000 may include an interface device 4006. The interface device may include one or more interface devices 4006. Interface device 4006 may include one or more communication chips, connectors, and / or other hardware and software to manage communication between computing device 4000 and other computing devices. For example, interface device 4006 may include circuitry for managing wireless communication of data transmission to and from computing device 4000. The term "wireless" and its derivatives can be used to describe circuits, devices, systems, methods, techniques, communication channels, etc., that can transmit data through a non-solid medium using modulated electromagnetic radiation. This term does not imply that the associated device does not contain any wires, although in some embodiments it may not contain any wires. The circuitry included in interface device 4006 for managing wireless communications may implement any of a number of wireless standards or protocols, including but not limited to Institute of Electrical and Electronics Engineers (IEEE) standards, including Wi-Fi (IEEE 802.11 series), IEEE 802.16 standards (e.g., IEEE 802.16-2005 amendments), the Long Term Evolution (LTE) project, and any amendments, updates, and / or revisions (e.g., Advanced LTE project, Ultra Mobile Broadband (UMB) project (also known as “3GPP2”), etc.). In some embodiments, the circuitry included in interface device 4006 for managing wireless communications may operate based on Global System for Mobile Communications (GSM), General Packet Radio Service (GPRS), Universal Mobile Telecommunications System (UMTS), High Speed ​​Packet Access (HSPA), Evolved HSPA (E-HSPA), or LTE networks. In some embodiments, the circuitry included in the interface device 4006 for managing wireless communications may operate based on Enhanced Data Rate GSM Evolution (EDGE), GSM EDGE Radio Access Network (GERAN), Universal Terrestrial Radio Access Network (UTRAN), or Evolved UTRAN (E-UTRAN). In some embodiments, the circuitry included in the interface device 4006 for managing wireless communications may operate based on Code Division Multiple Access (CDMA), Time Division Multiple Access (TDMA), Digital Enhanced Wireless Communication (DECT), Evolved Data Optimization (EV-DO) and its derivatives, as well as any other wireless protocol designated 3G, 4G, 5G, and higher. In some embodiments, the interface device 4006 may include one or more antennas (e.g., one or more antenna arrays) to receive and / or transmit wireless communications.

[0100] In some embodiments, interface device 4006 may include circuitry for managing wired communications, such as electrical communication protocols, optical communication protocols, or any other suitable communication protocols. For example, interface device 4006 may include circuitry supporting Ethernet-based communications. In some embodiments, interface device 4006 may support both wireless and wired communications, and / or may support multiple wired communication protocols and / or multiple wireless communication protocols. For example, a first set of circuitry for interface device 4006 may be dedicated to short-range wireless communications such as Wi-Fi or Bluetooth, and a second set of circuitry for interface device 4006 may be dedicated to long-range wireless communications such as Global Positioning System (GPS), EDGE, GPRS, CDMA, WiMAX, LTE, EV-DO, or others. In some embodiments, a first set of circuitry for interface device 4006 may be dedicated to wireless communications, and a second set of circuitry for interface device 4006 may be dedicated to wired communications.

[0101] The computing device 4000 may include a battery / power circuit 4008. The battery / power circuit 4008 may include one or more energy storage devices and / or circuitry for coupling components of the computing device 4000 to an energy source (e.g., AC line power) separate from the computing device 4000.

[0102] The computing device 4000 may include a display device 4010. The display device may include multiple display devices. The display device 4010 may include any visual indicator, such as a head-up display, computer monitor, projector, touch screen display, liquid crystal display (LCD), light-emitting diode display, or flat panel display.

[0103] The computing device 4000 may include other input / output (I / O) devices 4012. Other I / O devices 4012 may include, for example, one or more audio output devices (e.g., speakers, headphones, earphones, alarm clocks, etc.), one or more audio input devices (e.g., microphones or microphone arrays), positioning devices (e.g., GPS devices that communicate with satellite-based systems to receive the location of the computing device 4000, as known in the art), audio codecs, video codecs, printers, sensors (e.g., thermocouples or other temperature sensors, humidity sensors, pressure sensors, vibration sensors, accelerometers, gyroscopes, etc.), image capture devices (such as cameras), keyboards, cursor control devices (such as mice, styluses, trackballs, or touchpads), barcode readers, quick-response (QR) code readers, or radio frequency identification (RFID) readers.

[0104] The computing device 4000 may have any suitable form factor for its application and setup, such as a handheld or mobile computing device (e.g., a mobile phone, smartphone, mobile internet device, tablet computer, laptop computer, netbook computer, ultrabook computer, personal digital assistant (PDA), ultra-mobile personal computer, etc.), desktop computing device, server computing device, or other networked computing component.

[0105] One or more computing devices that implement any of the cell sorter support modules or methods disclosed herein may be part of a cell sorter support system. Figure 5 This is a block diagram of an exemplary support system 5000 based on various embodiments, which can be used to perform some or all of the cell sorting support methods disclosed herein. The cell sorting instrument support modules and methods disclosed herein include... Figures 1A-1C The cell sorter support modules 1100-1300 and the method 2000 in Figure 2 can be implemented by one or more components in the cell sorter support system 5000, such as cell sorter 5010, user local computing device 5020, service local computing device 5030 or remote computing device 5040.

[0106] Any cell sorter 5010, user local computing device 5020, service local computing device 5030, or remote computing device 5040 may contain references to this document. Figure 4 Any implementation of the computing device 4000 discussed herein, and any cell sorter 5010, user local computing device 5020, service local computing device 5030, or remote computing device 5040 may have the features referred to herein. Figure 4 The computing device 4000 under discussion may take any appropriate form.

[0107] Each of the cell sorter 5010, the user local computing device 5020, the service local computing device 5030, or the remote computing device 5040 may include a processing device 5002, a storage device 5004, and an interface device 5006. The processing device 5002 may take any suitable form, including those referenced herein. Figure 4 The processing device 5002 discussed herein may take any form, and the processing device 5002 contained in different cell sorters 5010, user local computing devices 5020, service local computing devices 5030, or remote computing devices 5040 may take the same or different forms. The storage device 5004 may take any suitable form, including those referenced herein. Figure 4Any form of storage device 5004 discussed herein, and including storage devices 5004 in different cell sorters 5010, user local computing devices 5020, service local computing devices 5030, or remote computing devices 5040, may take the same or different forms. Interface device 5006 may take any suitable form, including those referenced herein. Figure 4 The interface device 5006 discussed may take the same or different forms, and may be included in different cell sorters 5010, user local computing devices 5020, service local computing devices 5030 or remote computing devices 5040.

[0108] Cell sorter 5010, user local computing device 5020, service local computing device 5030, and remote computing device 5040 can communicate with other components of cell sorter support system 5000 via communication path 5008. As shown in the figure, communication path 5008 can connect the interface devices 5006 of different components in cell sorter support system 5000 for communication, and can be a wired or wireless communication path (e.g., as described herein). Figure 4 (The implementation of any communication technology discussed in the interface device 4006 of the computing device 4000). Figure 5 The specific cell sorter support system 5000 depicted includes communication paths between each pair of cell sorters 5010, user local computing device 5020, service local computing device 5030, and remote computing device 5040. However, this "fully connected" implementation is illustrative only, and various communication paths in communication path 5008 may not exist in different embodiments. For example, in some embodiments, there may not be a direct communication path 5008 between the service local computing device 5030 and its interface device 5006, or between the service local computing device 5030 and its interface device 5006 of the cell sorter 5010. Instead, communication with the cell sorter 5010 may occur via a communication path 5008 between the service local computing device 5030 and the user local computing device 5020, and between the user local computing device 5020 and the cell sorter 5010. The cell sorter 5010 can employ any suitable cell sorting device, such as BIGFOOT. TM Spectroscopic cell sorting instrument.

[0109] User local computing device 5020 may be a computing device located near the user to whom cell sorter 5010 belongs (e.g., any implementation of computing device 4000 discussed herein). In some embodiments, user local computing device 5020 may also be local to cell sorter 5010, but this is not necessarily the case; for example, user local computing device 5020 located in a user's home or office may be remote from cell sorter 5010 but communicate with it, allowing the user to use user local computing device 5020 to control and / or access data from cell sorter 5010. In some embodiments, user local computing device 5020 may be a laptop computer, smartphone, or tablet device. In some embodiments, user local computing device 5020 may be a portable computing device.

[0110] Figure 6A , 6B The 6C illustrates a typical workflow using a spectral cell sorting device, which processes the spectral feature data of the cell sorter via a neural network. (Example:) Figure 6A The workflow shown involves a cell passing through an array of linear lasers, generating fluorescence scattering from each laser. The fluorescence intensity is converted into a digital signal by the optical and PMT arrays of the cell sorter. This data is received by a network, and an onboard FPGA processes it via a neural network. In some embodiments, preprocessing operations such as scaling or spectral decomposition may be performed on the FPGA before the data enters the neural network, or they may be omitted. This information is then passed to software capable of plotting the generated two-dimensional output or representation onto the original visualization image. To classify cells based on these operations, the cell sorter must make a sorting decision within a mere 300 microseconds of laser detection. The FPGA-based architecture implemented in the cell sorter enables precise timing of mathematical operations, essential for high-throughput cell sorters. In some embodiments, this architecture can achieve processing speeds of up to 100,000 cells per second. In other embodiments, this architecture can achieve processing speeds exceeding 100,000 cells per second. Figure 6B The workflow shown begins with finding a non-parametric embedding. The neural network is then trained to reproduce the non-parametric embedding from the first cell sorter data. This is a typical supervised learning problem, solvable using optimizers like Adam to minimize mean squared error. After training, the neural network can be fed with data from the first cell sorter to generate a parametric embedding of that data. This parametric embedding can contain approximations of the original non-parametric embedding. Figure 6C The workflow shown involves directly training the network to generate parameterized embeddings. For example... Figure 6Band 6C The two workflows shown depict a user viewing a parametric embedding of data from the first cell sorter and specifying the necessary gating for sorting. The user can provide gating instructions, including specifying one or more sorting gatings for different classifications. These gatings can be determined by analyzing one or more of the measured fluorescence, light scattering data, fluorescent marker expression levels, and parametric embeddings. The neural network (including trained weights) and the gating can be stored in the instrument's FPGA for use in classifying cells corresponding to the second cell sorter data. In other words, the stored neural network and gating are used to classify another subset of cells in the sample that correspond to the second cell sorter data.

[0111] Figure 7A , 7B 7C and 7C respectively showcase excellent cell sorting data obtained using currently common analytical methods. Flow data are presented as a series of two-dimensional charts, including a set of exclusionary gating based on the expression levels of fluorescent markers to screen for desired cell populations.

[0112] Figure 8A and 8B Two plots are shown, in which CD4+ T cells are highlighted in both the tSNE and UMAP embeddings. Clearly, multiple distinct populations exist within the CD4+ subset. This might not be apparent in conventional screening methods. For example, in some cases, multiple distinct CD4+ cell populations may appear due to varying expression levels of other markers used in the two-dimensional plot. For instance, the single-detection mode used in conventional screening protocols can simultaneously display both CD4+ / CD8+ and CD4+ / CD8- cell populations.

[0113] Figures 9A to 9C This paper demonstrates a gated sequence formed using a gated localization algorithm, a typical algorithm that can be used in conjunction with the methods described in this disclosure. To provide relevant information about the content of each cluster, we developed a gated localization algorithm. Users can include any cluster in their embedding space and see a familiar set of gates that can distinguish the cells of interest.

[0114] Figures 10A to 10D The data obtained at each step of the method based on this disclosure are shown. Figure 10A and 10B In this study, a small subset of the fully mixed samples underwent a first UMAP analysis. The mapping was used to train a neural network to help it understand how this embedding mechanism works. Figure 10C and 10DIn this approach, a neural network is trained to generate a parameterized UMAP embedding graph without pre-determining the parameters of the UMAP embedding graph. A prototype network built using nine hidden layers (each with a size of 64) is able to reconstruct observed clustering and topological features with high accuracy, comparable to traditional UMAP methods. Using these two tools, users can pinpoint any cluster in their embedding space, gain a comprehensive understanding of that cluster, and incorporate new cell models into their embedding space in real time.

[0115] User local computing device 5020 may be a computing device located near the user to whom cell sorter 5010 belongs (e.g., any implementation of computing device 4000 discussed herein). In some embodiments, user local computing device 5020 may also be local to cell sorter 5010, but this is not necessarily the case; for example, user local computing device 5020 located in a user's home or office may be remote from cell sorter 5010 but communicate with it, allowing the user to use user local computing device 5020 to control and / or access data from cell sorter 5010. In some embodiments, user local computing device 5020 may be a laptop computer, smartphone, or tablet device. In some embodiments, user local computing device 5020 may be a portable computing device. In some embodiments, user local computing device 5020 may include one or more hardware components for performing computing operations, such as a central processing unit (CPU), graphics processing unit (GPU), tensor processing unit (TPU), or other suitable processor or computing device.

[0116] The service local computing device 5030 may be a computing device local to the entity serving the spectrometer 5010 (e.g., any embodiment of the computing device 4000 discussed herein). For example, the service local computing device 5030 may be local to the manufacturer of the cell sorter 5010 or a third-party service company. In some embodiments, the service local computing device 5030 may communicate with the cell sorter 5010, the user local computing device 5020, and / or the remote computing device 5040 (e.g., via a direct communication path 5008 or via multiple “indirect” communication paths 5008, as described above) to receive operational data about the cell sorter 5010, the user local computing device 5020, and / or the remote computing device 5040 (e.g., the results of the cell sorter 5010’s self-test, the calibration coefficients used by the cell sorter 5010, sensor measurements associated with the cell sorter 5010, etc.). In some embodiments, the service local computing device 5030 may communicate with the cell sorter 5010, the user local computing device 5020, and / or the remote computing device 5040, for example, via a direct communication path 5008 or via multiple "indirect" communication paths 5008 (as described above), to transmit data to the cell sorter 5010, the user local computing device 5020, and / or the remote computing device 5040, for example, to update programming instructions (such as firmware) in the cell sorter 5010, initiate test or calibration sequences in the cell sorter 5010, update programming instructions (such as software) in the user local computing device 5020 or the remote computing device 5040, and so on. Users of the cell sorter 5010 can use the cell sorter 5010 or the user local computing device 5020 to communicate with the service local computing device 5030 to report problems with the cell sorter 5010 or the user local computing device 5020, request technicians to improve the operation of the cell sorter 5010, order consumables or replacement parts related to the cell sorter 5010, or use it for other purposes.

[0117] Remote computing device 5040 may be a computing device located remotely from cell sorter 5010 and / or user local computing device 5020 (e.g., any embodiment of computing device 4000 discussed herein). In some embodiments, remote computing device 5040 may be included in a data center or other large-scale server environment. In some embodiments, remote computing device 5040 may include network-attached storage devices (e.g., as part of storage device 5004). Remote computing device 5040 may store data generated by cell sorter 5010, analyze data generated by cell sorter 5010 (e.g., according to preset instructions), facilitate communication between user local computing device 5020 and cell sorter 5010, and / or facilitate communication between service local computing device 5030 and cell sorter 5010.

[0118] In some embodiments, it may not exist. Figure 5 The cell sorter support system 5000 shown includes one or more components. Furthermore, in some embodiments, there may be... Figure 5 The cell sorting support system 5000 comprises multiple components among various elements. For example, the cell sorting support system 5000 may include multiple user local computing devices 5020 (e.g., different user local computing devices 5020 associated with different users or located in different locations). In another example, the cell sorting support system 5000 may include multiple cell sorters 5010, all of which communicate with a service local computing device 5030 and / or a remote computing device 5040; in such embodiments, the service local computing device 5030 may monitor these multiple cell sorters 5010, and the service local computing device 5030 may cause updates or other information to be simultaneously "broadcast" to multiple cell sorters 5010. Different cell sorters among the cell sorters 5010 in the cell sorting support system 5000 may be located near each other (e.g., in the same room) or far from each other (e.g., on different floors of a building, in different buildings, in different cities, etc.). In some embodiments, the cell sorter 5010 can be connected to an Internet of Things (IoT) stack that allows command and control of the cell sorter 5010 via web-based applications, virtual or augmented reality applications, mobile applications, and / or desktop applications. Any of these applications can be accessed by a user operating a user-local computing device 5020 that communicates with the cell sorter 5010 via an intermediary remote computing device 5040. In some embodiments, the manufacturer may sell the cell sorter 5010, along with one or more associated user-local computing devices 5020, as part of a local cell sorter computing unit 5012.

[0119] In some embodiments, the different types of cell sorters 5010 included in the cell sorting support system 5000 may be different sorters 5010; for example, one cell sorter 5010 may be a cell sorter, while another cell sorter 5010 may be an analyzer with cell sorting capabilities. For example, the cell sorter 5010 is equipped with a flow cytometer with cell sorting capabilities. The cell sorter may be a fluorescence activated cell sorting (FACS) system. In some such embodiments, a remote computing device 5040 and / or a user local computing device 5020 may combine data from the different types of cell sorters 5010 included in the cell sorter support system 5000.

[0120] The following paragraphs provide various examples of the embodiments disclosed herein.

[0121] Example 1 system includes: a cell sorting device, and a computing device configured to perform the following steps: collecting first cell sorter data associated with a first portion of one or more cell samples; training a machine learning model based on the first cell sorter data, user input, a representation of the first cell sorter data, or a combination thereof; receiving one or more sorting instructions from user input based on the first cell sorter data, the machine learning model, or a combination thereof; receiving second cell sorter data associated with a second portion of one or more cell samples; and determining a classification of the second cell sorter data, a representation of the second cell sorter data, or a combination thereof, based on the trained machine learning model and the sorting instructions input by the user; and

[0122] Based on the classification criteria, the samples were partially sorted.

[0123] Example 2 may include the subject matter covered in Example 1, and may further specify that the initial presentation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data.

[0124] Example 3 is a method for sorting particle samples, the method comprising: collecting first cell sorter data associated with a first portion of a particle sample containing one or more cells; training a machine learning model based on the first cell sorter data, user input, a representation of the first cell sorter data, or a combination thereof; receiving one or more sorting instructions from user input based on the first cell sorter data, the machine learning model, or a combination thereof; receiving second cell sorter data associated with a second portion of one or more cell samples; determining a classification of the second cell sorter data, a representation of the second cell sorter data, or a combination thereof based on the trained machine learning model and the sorting instructions input by the user; and sorting a portion of the samples based on the classification.

[0125] Example 4 may include the topics covered in Example 3, and may further specify that one or more gating instructions are determined based on the data output from the machine learning model, the embedded output of the machine learning model, the parameterized embedding of cell sorter data generated using the machine learning model, or a combination thereof.

[0126] Example 5 may include the subject matter of Example 3 or Example 4, and may further specify user input including gating instructions, sorting instructions, or combinations thereof, indicating one or more groups related to the data from the first cell sorter.

[0127] Example 6 may include the subject matter covered in Example 5, and may further specify the following: one or more groupings include one or more signal clustering or groupings with one or more parameters, said one or more parameters including: removal of duplicate cells, cell viability, light scattering, expression of one or more lineage markers, or one or more combinations of the above.

[0128] Example 7 may include the subject matter of any of Examples 3 through 6, and may further specify that the cell sorter includes a field-programmable gate array (FPGA), wherein the FPGA is programmed with parameters of a trained machine learning model and configured to analyze multiple portions of a sample using the trained machine learning model, and configured to store or include one or more trained weights of the trained machine learning model.

[0129] Example 8 may include the subject matter of any of Examples 3 to 7, and may further specify that the first representation of the first cell sorter data is determined based on a mapping process applied to the first cell sorter data, the mapping process including one or more processes, such as clustering, dimensionality reduction, embedding, nonparametric embedding or a combination thereof.

[0130] Example 9 may include the topics covered in Example 8, and may further specify that the embedding methods include Unified Manifold Approximation and Projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other nonlinear embedding methods, or combinations of the above methods.

[0131] Example 10 may include the topics covered in any of Examples 3 through 9, and may further illustrate that the machine learning model transforms high-dimensional cell sorter data into a low-dimensional representation of the transformed cell sorter data.

[0132] Example 11 may include the subject matter covered in Example 10, and may further specify that the transformed cell sorter data includes a low-dimensional representation that is optimized to preserve one or more features from the high-dimensional representation.

[0133] Example 12 may include the subject matter covered in Example 10, and may further specify the representation of cell sorter data, i.e., the representation includes one or more clusters of cell sorter data for separation into subsequent data representations.

[0134] Example 13 may include the subject matter covered in any of Examples 3 to 12, and may further specify that cell sorter data includes quantitative fluorescence data, which is represented by one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations of the above.

[0135] Example 14 may include the subject matter covered in Example 13, and may further specify that quantitative fluorescence data includes one or more fluorescence signals.

[0136] Example 15 may include the subject matter covered in Example 13, and may further specify that the fluorescent signal originates from one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, one or more groups of fluorescent beads or fluorescently labeled beads, or one or more combinations of the foregoing.

[0137] Example 16 may include the subject matter covered in any of Examples 3 through 15, and may further specify that the machine learning model includes a neural network.

[0138] Example 17 may include the topics covered in Example 3, and may further clarify that the training process of the machine learning model does not require prior determination of a first representation of the first cell sorter data.

[0139] Example 18 may include the subject matter covered in Example 16, and may further specify that the neural network is trained to learn a mapping process mechanism for generating one or more representations of cell sorter data in a cell sorter measurement session.

[0140] Example 19 may include the subject matter covered in Example 16, and may further specify that the neural network includes one or more artificial neural networks, convolutional neural networks, recurrent neural networks, or any combination of the above.

[0141] Example 20 may include the subject matter covered in any of Examples 16, 18, or 19, and may further specify that the neural network contains one or more non-linear activation functions.

[0142] Example 21 may include the topics covered in Example 20, and may further specify that these nonlinear activation functions include, but are not limited to, the modified linear unit (ReLU), the leaky ReLU function, the hard sigmoid function, or one or more combinations thereof.

[0143] Example 22 may include the subject matter covered in any of Examples 16, 18 or 19, and may further specify that the neural network includes a series of affine transformations and a nonlinear activation function for the first cell sorter data.

[0144] Example 23 may include the topics covered in any of Examples 16 through 22, and may further illustrate that the neural network processes cell sorter data and provides output in approximately 300 microseconds.

[0145] Example 24 may include the subject matter covered by any of Examples 3 to 23, and may further specify the representation of the first cell sorter data and the representation of the second cell sorter data, wherein each of the two representations includes at least one corresponding two-dimensional or three-dimensional plot.

[0146] Example 25 may include the subject matter of any of Examples 3 through 24, and may further specify that the determination process of the second cell sorter data representation includes processing the machine learning model at a rate of approximately 100,000 events per second or higher.

[0147] Example 26 may include the subject matter covered by any of Examples 3 to 25, and may further specify that the sample includes a biological sample containing multiple cells.

[0148] Example 27 may include the subject matter covered in any of Examples 3 through 26, and may further specify that the classification is based on one or more cell phenotypic markers.

[0149] Example 28 may include the subject matter covered in any of Examples 3 through 27, and may further specify that the cell sorter includes one or more processors, which at least include a central processing unit (CPU), a graphics processing unit (GPU), and a field-programmable gate array (FPGA), wherein these processors pass the classification results of the second cell sorter data to the cell sorter in order to sort a portion of the sample.

[0150] Example 29 is a method for analyzing a sample containing multiple particles, the method comprising: receiving cell sorter data associated with one or more sensors of an instrument; determining a representation of the cell sorter data associated with a mapping process based on an operation of inputting the cell sorter data into a machine learning model; and outputting the representation of the cell sorter data.

[0151] Example 30 may include the subject matter covered in Example 29, and may further specify that the steps leading to the output include one or more of the following operations: displaying a representation of cell sorter data, sending a representation of cell sorter data to a computing device via a network, causing the user interface to update to control the cell sorter to present the measurement results, or a combination of the above operations.

[0152] Example 31 may include the subject matter covered in Example 29 or 30, and may further clarify the steps of receiving user input based on the output representation of cell sorter data, and sorting multiple particles based on the user input.

[0153] Example 32 may include the topics covered in Example 29 or 30, and may further specify that user input includes setting a gating threshold based on cell sorter data.

[0154] Example 33 is a method for classifying particle samples, the method comprising: training a machine learning model based on one of the following: first cell sorter data, user input, a representation of the first cell sorter data, or a combination thereof; receiving second cell sorter data associated with a second portion of cells in the sample; and sorting a portion of the sample based on a classification determined from the trained machine learning model and / or a gating instruction provided by the user.

[0155] Example 34 may include the subject matter covered in Example 3, and may further specify that the first representation of the first cell sorter data is determined based on a mapping process applied to the first cell sorter data, the mapping process including one or more processes, such as clustering, dimensionality reduction, embedding, nonparametric embedding, or a combination thereof.

[0156] Example 35 is a method for analyzing a particle sample, the method comprising: collecting first cell sorter data associated with a first portion of a sample consisting of one or more cells; training a machine learning model based on the first cell sorter data, user input, a representation of the first cell sorter data, or a combination thereof; determining a parameterized embedding of the first cell sorter data based on first measurement data and the trained machine model; and receiving one or more gating instructions from a user using the first cell sorter data, the parameterized embedding of the first cell sorter data, or a combination thereof.

[0157] Receive second cell sorter data associated with a second portion of one or more cell samples; determine the embedding coordinates of the second cell sorter data using a trained machine learning model; determine the classification of the second cell sorter data based on the embedding coordinates of the second cell sorter data and user-provided gating instructions; and sort the second portion of samples based on the classification.

[0158] Example 36 may include the subject matter covered in Example 35, and may further specify that the first representation of the first cell sorter data is determined based on a mapping process applied to the first cell sorter data, the mapping process including one or more processes, such as clustering, dimensionality reduction, embedding, nonparametric embedding, or a combination thereof.

[0159] Example 37 is a method for analyzing particle samples, the method comprising: determining one or more configuration parameters for training machine learning of a cell sorter based on user input; receiving first cell sorter data from the cell sorter; training a representation of the first cell sorter data or one or more combinations thereof, based on the first cell sorter data and the user input, to generate a machine learning model; and storing the machine learning model, wherein a computer processor communicating with the cell sorter is configured to use the stored machine learning model to analyze second cell sorter data from the cell sorter.

[0160] Example 38 may include the topics covered in Example 37, and may further clarify that the machine learning model includes a neural network.

[0161] Example 39 may include the subject matter covered in Example 38, and may further specify that a neural network includes one or more activation functions, including but not limited to the rectified linear unit (ReLU), the leaky ReLU function, the hard sigmoid function, or one or more combinations thereof.

[0162] Example 40 may include the subject matter of any of Examples 37 to 39, and may further specify that the cell sorter data includes microscopic data, hyperspectral imaging data, high-dimensional vector data, or one or more combinations thereof.

[0163] Example 41 may include the subject matter covered in any of Examples 37 to 40, and may further specify that cell sorter data includes quantitative fluorescence data, which is represented by one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations of the above.

[0164] Example 42 may include the subject matter covered in Example 41, and may further specify that quantitative fluorescence data includes one or more fluorescence signals from one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, one or more fluorescent beads or groups of fluorescently labeled beads, or one or more combinations of the foregoing.

[0165] Example 43 may include the subject matter covered in any of Examples 37 to 42, and may further specify that the first representation of the first cell sorter data is determined based on a mapping process applied to the first cell sorter data, the mapping process including one or more processes, such as clustering, dimensionality reduction, embedding, nonparametric embedding, or a combination thereof.

[0166] Example 44 may include the subject matter covered in Example 43, and may further specify that the embedding method includes Unified Manifold Approximation and Projection (UMAP), t-distributed random neighborhood embedding (t-SNE), other nonlinear embedding methods, or combinations of the above methods.

[0167] Example 45 may include the topics covered in any of Examples 37 through 44, and may further illustrate that the machine learning model transforms high-dimensional cell sorter data into a low-dimensional representation of the transformed cell sorter data.

[0168] Example 46 may include the subject matter covered by any of Examples 37 to 45, and may further specify the representation of the first cell sorter data and the representation of the second cell sorter data, wherein each of the two representations includes at least one corresponding two-dimensional or three-dimensional plot.

[0169] Example 47 may include the subject matter of any of Examples 37 to 46, and may further specify that determining the representation of the second cell sorter data includes sorting the second cell sorter data using a field-programmable gate array optimized for real-time sorting decisions on the cell sorter.

[0170] Example 48 may include the subject matter of any of Examples 37 to 47, and may further specify that the determination process of the second cell sorter data representation includes processing the machine learning model at a rate of approximately 100,000 events per second or higher.

[0171] Example 49 is a method for displaying flow cytometry data in real time, the method comprising: outputting a representation of a first portion of a sample via a user interface associated with a cell sorter; receiving instructions via the user interface for generating a machine learning model of the sample, wherein the machine learning model is trained based on a mapping process that maps cell sorter data to an embedding space; receiving instructions via the user interface for processing a second portion of a sample, wherein the instruction process includes instructions for sorting the sample using the machine learning model; and outputting a representation of second cell sorter data via the user interface based on the machine learning model.

[0172] Example 50 device includes: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the device to perform any of the methods in Examples 3 to 49.

[0173] Example 51 is a non-transitory computer-readable medium that stores instructions that, when executed by one or more processors, cause a cell sorter to perform the operations of any of the methods in Examples 3 to 49.

[0174] Example 52 is a system for analyzing multiple particles, comprising: a cell sorter comprising: one or more sensors configured to generate cell sorter data; and a computing device comprising one or more processors and a memory, wherein the memory stores instructions that, when executed by the one or more processors, cause the computing device to perform any of the methods in Examples 3 through 49.

Claims

1. A system comprising: Cell sorting equipment, and A computing device configured to perform the following steps: Collect first cell sorter data associated with a first subset of cells in one or more cell samples; Construct a machine learning model based on one or more first cell sorter data, user input, a certain representation of the first cell sorter data (or a combination thereof), or any combination of the above. Based on data from the first cell sorter, a machine learning model, or a combination of both, it receives one or more sorting instructions from user input. Receive second cell sorter data associated with a second portion of one or more cell samples; Based on a trained machine learning model and user-input gating commands, determine the classification of the second cell sorter data, the representation of the second cell sorter data, or a combination of both. as well as Based on the classification criteria, the samples were partially sorted.

2. The system of claim 1, wherein the first representation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data.

3. A method for sorting particle samples, the method comprising: Collect cell sorter data from the first portion of a sample consisting of one or more particles; Construct a machine learning model based on one or more first cell sorter data, user input, a certain representation of the first cell sorter data (or a combination thereof), or any combination of the above. Based on data from the first cell sorter, a machine learning model, or a combination of both, it receives one or more sorting instructions from user input. Receive second cell sorter data associated with a second portion of one or more cell samples; Based on the trained machine learning model and the gating instructions input by the user, the classification of the second cell sorter data, the representation of the second cell sorter data, or a combination of both are determined; then, the samples are sorted based on the classification criteria, and a portion of the samples are selected.

4. The method of claim 3, wherein the one or more gating instructions are determined based on data output from a machine learning model, embedded output from a machine learning model, parameterized embedding of cell sorter data generated using a machine learning model, or a combination thereof.

5. The method of claim 3 or 4, wherein the user input includes gating instructions, sorting instructions, or combinations thereof, indicating one or more groups associated with the first cell sorter data.

6. The method of claim 5, wherein the one or more packets comprise one or more signal packets or gates for one or more parameters, the parameters including: Remove the expression of duplicate cells, cell viability, light scattering, one or more lineage markers, or one or more combinations of the above.

7. The method of any one of claims 3 to 6, wherein the cell sorter includes a field-programmable gate array (FPGA), wherein the FPGA is programmed with parameters of a trained machine learning model and configured to analyze multiple portions of a sample using the trained machine learning model, and configured to store or include one or more trained weights of the trained machine learning model.

8. The method of any one of claims 3 to 7, wherein the first representation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data, the mapping process including one or more clustering processes, dimensionality reduction processes, embedding, non-parametric embedding or combinations thereof.

9. The method of claim 8, wherein the embedding comprises uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), another nonlinear embedding, or a combination thereof.

10. The method of any one of claims 3 to 9, wherein the machine learning model converts high-dimensional cell sorter data into a representation of converted cell sorter data with lower dimensions.

11. The method of claim 10, wherein the converted cell sorter data comprises an optimized low-dimensional representation designed to preserve one or more features of the high-dimensional representation.

12. The method of claim 10, wherein the representation of cell sorter data comprises one or more clusters of cell sorter data for separation into subsequent data representations.

13. In the method of any one of claims 3 to 12, the cell sorter data includes quantitative fluorescence data, which represents one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations thereof.

14. The method of claim 13, wherein the quantitative fluorescence data comprises one or more fluorescence signals.

15. The method of claim 13, wherein the fluorescence signal originates from one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, one or more groups of fluorescent beads or fluorescently labeled microbeads, or a combination thereof.

16. The method of any one of claims 3 to 15, wherein the machine learning model comprises a neural network.

17. The method of claim 3, wherein the training process of the machine learning model does not involve pre-determining the initial representation of the first cell sorter data.

18. The method of claim 16, wherein the neural network is trained to learn a mapping process mechanism for generating one or more representations of cell sorter data in a cell sorter measurement session.

19. The method of claim 16, wherein the neural network comprises one or more artificial neural networks, convolutional neural networks, recurrent neural networks, or one or more combinations thereof.

20. The method of any one of claims 16, 18 or 19, wherein the neural network comprises one or more nonlinear activation functions.

21. The method of claim 20, wherein the one or more nonlinear activation functions include, but are not limited to: modified linear unit (ReLU), leaky ReLU function, hard sigmoid function, or any combination of the above functions.

22. The method of any one of claims 16, 18 or 19, wherein the neural network comprises one or more affine transformation sequences with respect to the first cell sorter data and a nonlinear activation function.

23. The method of any one of claims 16 to 22, wherein the neural network processes cell sorter data and provides output in about 300 microseconds.

24. The method of any one of claims 3 to 23, wherein the representation of the first cell sorter data and the representation of the second cell sorter data each comprise at least one corresponding two-dimensional or three-dimensional graph.

25. The method of any one of claims 3 to 24, wherein determining the representation of the second cell sorter data comprises processing the machine learning model at a frequency of about 100,000 events per second or higher.

26. The method of any one of claims 3 to 25, wherein the sample comprises a biological sample containing a plurality of cells.

27. The method of any one of claims 3 to 26, wherein the classification is based on one or more cell phenotypic markers.

28. The method of any one of claims 3 to 27, wherein the cell sorter includes one or more processors, the processors including at least a central processing unit (CPU), a graphics processing unit (GPU), and a field-programmable gate array (FPGA), wherein the one or more processors transmit the classification results of the second cell sorter data to the cell sorter to sort a portion of the sample.

29. A method for analyzing a sample containing multiple particles, the method comprising: Receive cell sorting data associated with one or more sensors of the instrument; By inputting cell sorter data into a machine learning model, the representation of cell sorter data related to the mapping process is determined; It also outputs a representation of the cell sorter data.

30. The method of claim 29, wherein the output-causing step may include one or more of the following operations: displaying a representation of cell sorter data, sending the representation of cell sorter data to a computing device via a network, causing an update to a user interface for controlling the cell sorter to present measurement results, or a combination of the above operations.

31. The method of claim 29 or 30 further includes receiving user input based on the output representation of cell sorting data, and sorting a plurality of particles based on the user input.

32. The method of claim 29 or 30, wherein the user input includes setting a gating threshold based on cell sorter data.

33. A method for sorting particle samples, the method comprising: Construct a machine learning model based on one or more first cell sorter data, user input, a certain representation of the first cell sorter data (or a combination thereof), or any combination of the above. Receive second cell sorter data associated with a second portion of one or more cell samples; and A portion of the samples are sorted based on classification results determined by a trained machine learning model and / or filtering instructions provided by the user.

34. The method of claim 33, wherein the first representation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data, the mapping process including one or more clustering processes, dimensionality reduction processes, embedding, non-parametric embedding, or combinations thereof.

35. A method for analyzing particle samples, the method comprising: Collect cell sorter data from the first portion of a sample consisting of one or more particles; Construct a machine learning model based on one or more first cell sorter data, user input, a certain representation of the first cell sorter data (or a combination thereof), or any combination of the above. Based on the first measurement data and the trained machine model, the first cell sorter data is parametrically embedded. Based on one or more filtering instructions provided by the user, these instructions can be obtained based on the first cell sorter data, the parameterized embedding of the first cell sorter data, or a combination of both; Receive second cell sorter data associated with a second portion of one or more cell samples; The coordinate information from the second cell sorter data is embedded using a trained machine learning model. The data from the second cell sorter is classified based on the embedded coordinates of the data and the gating commands provided by the user. and Based on the classification criteria, a partial screening was performed on the second part of the sample.

36. The method of claim 35, wherein a first representation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data, the mapping process including one or more clustering processes, dimensionality reduction processes, embedding, non-parametric embedding, or combinations thereof.

37. A method for analyzing particle samples, the method comprising: Based on user input, determine one or more configuration parameters for training the cell sorter machine learning model; Receive the first cell sorting data from the cell sorter; A machine learning model is constructed by training based on data from the first cell sorter, user input, a certain representation of the first cell sorter data (or a combination thereof), or any combination of the above. The system stores machine learning models, wherein a computer processor communicating with a cell sorter is configured to use the stored machine learning models to analyze second cell sorting data from the cell sorter.

38. The method of claim 37, wherein the machine learning model comprises a convolutional neural network.

39. The method of claim 38, wherein the neural network comprises one or more activation functions, including but not limited to the modified linear unit (ReLU), the leaky ReLU function, the hard sigmoid function, or one or more combinations thereof.

40. The method of any one of claims 37 to 39, wherein the cell sorter data includes microscopic data, hyperspectral imaging data, high-dimensional vector data, or one or more combinations thereof.

41. In the method of any one of claims 37 to 40, the cell sorter data includes quantitative fluorescence data, which represents one or more antibodies bound to each cell, antibody binding capacity (ABC), equivalent soluble fluorescent dye molecule (MESF), one or more other quantitative fluorescence indicators, or one or more combinations thereof.

42. The method of claim 41, wherein the quantitative fluorescence data comprises one or more fluorescence signals, one or more fluorescent proteins, one or more fluorescent dyes, one or more fluorescently labeled antibodies, one or more fluorescent beads or a group of fluorescently labeled beads, or one or more combinations thereof.

43. The method of any one of claims 37 to 42, wherein a first representation of the first cell sorter data is determined by applying a mapping process to the first cell sorter data, the mapping process including one or more clustering processes, dimensionality reduction processes, embedding, non-parametric embedding, or combinations thereof.

44. The method of claim 43, wherein the embedding comprises uniform manifold approximation and projection (UMAP), t-distributed random neighborhood embedding (t-SNE), another nonlinear embedding, or a combination thereof.

45. The method of any one of claims 37 to 44, wherein the machine learning model converts high-dimensional cell sorter data into a representation of the converted cell sorter data with lower dimensions.

46. ​​The method of any one of claims 37 to 45, wherein the representation of the first cell sorter data and the representation of the second cell sorter data each comprise at least one corresponding two-dimensional or three-dimensional graph.

47. The method of any one of claims 37 to 46, wherein determining the representation of the second cell sorter data comprises sorting the second cell sorter data using a field-programmable gate array optimized for making real-time sorting decisions on a cell sorter.

48. The method of any one of claims 37 to 47, wherein determining the representation of the second cell sorter data comprises processing the machine learning model at a frequency of about 100,000 events per second or higher.

49. A method for real-time display of flow cytometry data, the method comprising: The user interface associated with the cell sorter outputs an image representation of the first part of the sample; The user interface receives instructions to generate a machine learning model for samples, which is trained based on a mapping process that maps cell sorter data to an embedding space. The second part of the instruction to process the sample is received through the user interface, wherein the processing of the instruction includes instructions to classify the sample using a machine learning model; and The second cell sorter data is visualized through the user interface and based on a machine learning model.

50. An apparatus comprising: One or more processors; and A memory storing instructions, which, when executed by one or more processors, causes the device to perform the method of any one of embodiments 3 to 49.

51. A non-transitory computer-readable medium storing instructions that, when executed by one or more processors, cause a computing system to perform the method according to any one of claims 3 to 49.

52. A system for analyzing multiple particles, comprising: A cell sorter comprises: one or more sensors configured to generate cell sorter data; and a computing device including one or more processors and a memory, wherein the memory stores instructions that, when executed by the one or more processors, cause the computing device to perform any of the methods in Examples 3 to 49.

Images