Single cell identification for cell sorting

Abstract

The single-cell identification described here utilizes cell image information and uses a neural network model to extract cellular features to subtly distinguish noise events from single cells, allowing users to select which different types of noise events should be filtered out depending on the application requirements. Because high-speed neural network models can extract richer and more specific cellular features than manually selected features, the models can distinguish noise events and identify single cells in real time with higher accuracy and discrimination power. The use of neural network models for real-time single-cell identification is a novel technique that has not been applied before. It enables higher discrimination power and accuracy compared to conventional FACS (fluorescence-activated cell sorting). The utility of this technique lies in its integration with optional bright-field (BF) and fluorescence (FL) models to identify single cells for different downstream applications.

Description

[Technical Field]

[0001] The present invention relates to cell sorting, and more particularly to image-based cell sorting. [Background technology]

[0002] Conventional fluorescence-activated cell sorting (FACS) does not have image information. FACS uses the laser scattering signal of cells to detect noise events. FACS cannot distinguish between specific noise events that affect the accuracy of the experiment and the purity of the sorted product. Summary of the Invention [Problem to be solved by the invention]

[0003] The single-cell identification described herein utilizes cell image information and uses a neural network model to extract cellular features to subtly distinguish noise events from single cells, allowing users to select which different types of noise events to filter out depending on the application requirements. Because high-speed neural network models can extract richer and more specific cellular features than manually selected features, the models can provide higher accuracy and discrimination capabilities to distinguish noise events and identify single cells in real time. The use of neural network models for real-time single-cell identification is a novel technology that has not been applied before. It enables higher discrimination capabilities and accuracy compared to conventional FACS (fluorescence-activated cell sorting). The utility of this technology is its integration with optional bright-field (BF) and fluorescence (FL) models to identify single cells for different downstream applications. [Means for solving the problem]

[0004] In one aspect, a method programmed into a non-transitory memory of a device includes receiving an input, processing the input using a neural network to identify noise events or single cells, and classifying the input as the noise event or the single cell. The input includes a plurality of cell images. The neural network includes two convolutional layers and a fully connected layer. The neural network includes a plurality of kernels for extracting features from the input, including different fine-grained levels. The noise events are selected from the group consisting of aggregates, debris, doublets, edges, defocus, detached blebs, and blebs. Classifying the input includes labeling each image as the noise event or the single cell. The neural network is configured to classify more than 2,000 images per second. The method further includes implementing a bright-field model or a fluorescence model in parallel with classifying the input.

[0005] In another aspect, an apparatus includes: a non-transitory memory for storing an application, the application for receiving an input, processing the input using a neural network to identify a noise event or a single cell, and classifying the input as the noise event or the single cell; and a processor coupled to the memory and configured to process the application. The input includes a plurality of cell images. The neural network includes two convolutional layers and a fully connected layer. The neural network includes a plurality of kernels for extracting features from the input with different levels of granularity. The noise events are selected from the group consisting of aggregates, debris, doublets, edges, defocus, isolated blebs, and blebs. Classifying the input includes labeling each image with the noise event or the single cell. The neural network is configured to classify more than 2,000 images per second. The application is further configured to implement a bright-field model or a fluorescence model in parallel with classifying the input.

[0006] In another aspect, a system includes a first device configured to acquire a cell image and a second device configured to receive the cell image, process the cell image using a neural network to identify noise events or single cells, and classify the cell image as the noise event or the single cell. The neural network includes two convolutional layers and a fully connected layer. The neural network includes multiple kernels for extracting features with different levels of granularity from the cell image. The noise events are selected from the group consisting of aggregates, debris, doublets, edges, defocus, isolated blebs, and blebs. Classifying the cell image includes labeling the cell image as the noise event or the single cell. The neural network is configured to classify more than 2,000 images per second. The second device is further configured to implement a bright-field model or a fluorescence model in parallel with classifying the cell image. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 illustrates an image of an exemplary noise event according to some embodiments. [Figure 2] FIG. 1 is a diagram of a neural network architecture configured to implement single cell identification according to some embodiments. [Figure 3] 1 is a flowchart of a method for implementing single cell identification for cell sorting according to some embodiments. [Figure 4] FIG. 1 is a block diagram of an exemplary computing device configured to implement a single cell identification method according to some embodiments. DETAILED DESCRIPTION OF THE INVENTION

[0008] Flow cytometer samples contain desired single cells in suspension along with noise events such as aggregates, doublets, and debris. Additionally, some cells arrive off-center, resulting in cells that are out of focus or off the edge of the interrogation region. Finally, some cells exhibit blebbing (both attached and detached), which may indicate cellular stress and damage. Conventional flow cytometry-based cell sorters lack image information and are therefore limited in their ability to filter out noise events from single cells within a sample. Conventional cell sorting uses the cell's laser scattering signal to filter out distinct noise cells (doublets, clumps, and debris) but cannot distinguish between other noise events (edges, out-of-focus, blebs, and detached blebs). This impacts the accuracy of the experiment and the purity of the sorted population. The single cell identification described herein utilizes cell image information and uses neural network models to extract cell features to subtly distinguish noise events from single cells, allowing the user to select which different types of noise events to filter out depending on the application requirements.

[0009] The fast neural network models described herein are able to extract richer and more specific cellular features than manually selected features, allowing the models to distinguish noise events and identify single cells in real time with greater accuracy and discrimination.

[0010] The use of neural network models for real-time single-cell identification is a novel technique that has not been applied before. Compared to conventional FACS (fluorescence-activated cell sorting), it enables high discrimination ability and high accuracy. The utility of this technique is that it can be integrated with any bright-field (BF) model and fluorescence (FL) model to identify single cells for different downstream applications.

[0011] 1 shows an image of exemplary noise events according to some embodiments. The noise events include aggregates 100, doublets 102, debris 104, edges 106, out-of-focus 108, detached blebs 110, and blebs 112, as opposed to single cells 114. Aggregates 100 contain multiple cells, while doublets 102 contain two cells. Debris 104 is an artifact that is not a single cell. Cell edges 106 may be missing or the cells may be out-of-focus 108. Detachment blebs 110 are detached bubbles, and blebs 112 are attached bubbles or rounded protrusions on the surface of a cell.

[0012] Traditional FACS does not possess image information and can only filter out a subset of noise events present in all samples. Because FACS detects noise events using the cell's laser scattering signal, it can distinguish aggregates, doublets, and debris from single cells. FACS cannot distinguish single cells from other noise events (e.g., blebs, separation blebs, out-of-focus events, and edges). This impacts the accuracy of the experiment and the purity of the sorted product. In other words, FACS can only determine whether an image contains a single cell in a binary way. In contrast, in the single-cell identification implementation described herein, different noise classes exist. Depending on the application, a user may want to allow images containing specific noise classes. For example, if a user is sorting rare events and wants to select as many cells of interest as possible, the user may want to include images containing bleb events, edge events, and slightly out-of-focus events in the sorting because they do not want to filter out potentially useful information. Images can be further classified based on noise events, which can also limit or expand further processing based on the noise events.

[0013] For example, Table 1 shows the differences between FACS and potential single cell discrimination implementations. [Table 1]

[0014] FIG. 2 illustrates a diagram of a neural network architecture configured to implement single-cell identification according to some embodiments. The neural network 200 includes a module with two convolutional layers and one fully connected layer. The neural network 200 includes multiple kernels for extracting rich features, including different levels of granularity. The neural network 200 can process over 2,000 images per second using a 2080Ti GPU. The neural network 200 is highly accurate. When excluding 5C noise events on the publicly available WBC2020 dataset, the precision rate for single-cell identification is 98.6% and the recall rate is 97.3%. The neural network 200 can be integrated with a cell type classifier module. While a specific exemplary neural network architecture is described, the neural network 200 is not limited to that particular architecture. For example, the neural network 200 can include fewer, additional, or different components (e.g., fewer convolutional layers, multiple GPUs, or different GPUs).

[0015] The neural network 200 receives input (e.g., a large number of cell images) and uses machine learning to process the input to extract features as described herein and generate output, such as labeling each image with one or more noise events or as a single cell. The neural network can be trained using multiple data sets.

[0016] Neural network 200 can be implemented in BF applications by parallelizing with BF cell type classifiers by sharing feature extractors. Neural network 200 can be implemented in FL applications by parallelizing with FL cell type classifiers on a GPU. Neural network 200 can also be used in conjunction with cell sorting applications such as nuclear transport, immunoflowfish, and extracellular vesicles.

[0017] 3 shows a flowchart of a method for implementing single cell identification for cell sorting according to some embodiments. At step 300, input is received at the device. The input may include cell images and / or other content.

[0018] In step 302, a neural network is used to process the input. The neural network includes two convolutional layers and one fully connected layer, and multiple kernels extract features from the input with different levels of granularity. Processing the input includes using machine learning to distinguish noise and identify noise events. For example, the neural network learns discrimination factors such as detecting more than two circular objects to identify clusters, detecting spots and lines that are not part of cells to identify debris, detecting two circular objects to identify doublets, detecting nearly circular objects with parts of the circle missing to identify edges, detecting blurred or distorted lines to identify out-of-focus, detecting separated circles smaller than cells to identify separated blebs, and detecting attached circles smaller than cells to identify blebs. The neural network can also learn to identify a single cell if no noise events are detected.

[0019] In step 304, the input is classified. For example, the input is classified as having a noise event (e.g., agglomerates, debris, doublets, edges, defocus, isolated blebs, blebs) or a single cell. For example, each image of the input is labeled with a noise event or as a single cell. In some embodiments, if multiple noise events are detected, the image is classified and / or labeled with both noise events. In some embodiments, if multiple noise events are detected, the image is classified and / or labeled with the more prominent noise event as determined by a machine language.

[0020] This method can be implemented using a 2080Ti GPU to classify over 2,000 images per second with precision rates exceeding 98.6% and recall rates exceeding 97.3% for single-cell identification when excluding 5C noise events.

[0021] In some embodiments, fewer or additional steps are implemented. For example, a method for implementing single cell identification can be integrated with a BF application or an FL application. In another example, a neural network is trained using multiple datasets before additional inputs are processed. In some embodiments, the order of steps is changed.

[0022] FIG. 4 illustrates a block diagram of an exemplary computing device configured to implement a single-cell identification method according to some embodiments. The computing device 400 can be used to acquire, store, compute, process, communicate, and / or display information, such as images and videos, including 3D content. The computing device 400 can implement any aspect of single-cell identification. In general, a hardware configuration suitable for implementing the computing device 400 includes a network interface 402, memory 404, a processor 406, I / O device(s) 408, a bus 410, and storage 412. The choice of processor is not critical as long as a suitable processor with sufficient speed is selected. The processor 406 can include one or more central processing units (CPUs) and / or one or more graphics processing units (GPUs), such as a 2080Ti GPU. The memory 404 can be any conventional computer memory known in the art. The storage device 412 may include a hard drive, CD-ROM, CDRW, DVD, DVDRW, high-definition disk / drive, ultra-high-definition drive, flash memory card, or any other storage device. The computing device 400 may include one or more network interfaces 402. Examples of network interfaces include a network card connected to an Ethernet or other type of LAN. The I / O device(s) 408 may include one or more of a keyboard, mouse, monitor, screen, printer, modem, touch screen, button interface, and other devices. The single cell identification application(s) 430 used to implement the single cell identification implementation are likely stored in the storage device 412 and memory 404 and processed as applications are typically processed. The computing device 400 may include more or fewer components than those shown in FIG. 4 . In some embodiments, single cell identification hardware 420 is included.4 includes hardware 420 and application 430 for implementing single cell identification, the single cell identification method can be implemented in the computing device as hardware, firmware, software, or any combination thereof. For example, in some embodiments, the single cell identification application 430 is programmed into memory and executed using a processor. As another example, in some embodiments, the single cell identification hardware 420 is programmed hardware logic that includes gates specifically designed to implement the single cell identification method.

[0023] In some embodiments, the single cell identification application(s) 430 include several applications and / or modules. In some embodiments, a module also includes one or more sub-modules. In some embodiments, fewer or additional modules may be included.

[0024] Examples of suitable computing devices include a personal computer, a laptop computer, a computer workstation, a server, a mainframe computer, a handheld computer, a personal digital assistant, a cellular / mobile phone, a smart appliance, a game console, a digital camera, a digital camcorder, a camera phone, a smartphone, a portable music player, a tablet computer, a mobile device, a video player, a video disc writer / player (e.g., a DVD writer / player, a high-definition disc writer / player, an ultra-high-definition disc writer / player), a television, a home entertainment system, an augmented reality device, a virtual reality device, smart jewelry (e.g., a smart watch), a vehicle (e.g., an autonomous vehicle), or any other suitable computing device.

[0025] In some embodiments, a first computing device (e.g., a microscope / camera configuration) captures cell images and communicates the images to a second computing device that implements a single cell identification method to process and classify the images.

[0026] To utilize the single cell identification method, the device acquires or receives images of the cells to be sorted. The single cell identification method can be implemented with user assistance or automatically without user involvement.

[0027] In operation, the single cell identification described herein can distinguish single cells from any number (e.g., 7) types of noise events, allowing the user to select which different types of noise events to filter out depending on the requirements of the application. Single cell identification can be implemented as a neural network-based single cell identification module.

[0028] The neural network-based module extracts features that are superior to manually selected features for single-cell identification and achieves superior performance.

[0029] The high-speed single-cell identification module can be integrated with any BF or FL downstream application.

[0030] Some embodiments of single cell identification for cell sorting 1. A method programmed into the non-transitory memory of a device, comprising: receiving an input; processing the input using a neural network to identify noise events or single cells; classifying the input as the noise event or the single cell; A method comprising:

[0031] 2. The method of claim 1, wherein the input includes a plurality of cell images.

[0032] 3. The method of claim 1, wherein the neural network includes two convolutional layers and a fully connected layer.

[0033] 4. The method of claim 1, wherein the neural network includes multiple kernels for extracting features from the input with different levels of granularity.

[0034] 5. The method of claim 1, wherein the noise events are selected from the group consisting of aggregates, debris, doublets, edges, defocus, separation blebs, and blebs.

[0035] 6. The method of claim 1, wherein the step of classifying the input includes labeling each image as the noise event or the single cell.

[0036] 7. The method of claim 1, wherein the neural network is configured to classify more than 2,000 images per second.

[0037] 8. The method of claim 1, further comprising the step of implementing a bright field model or a fluorescence model in parallel with the step of classifying the input.

[0038] 9. An apparatus comprising: A non-transitory memory for storing an application, said application comprising: Receiving input; processing the input using a neural network to identify noise events or single cells; classifying the input as the noise event or the single cell; a non-transitory memory for performing a processor coupled to the memory and configured to process the application; An apparatus comprising:

[0039] 10. The apparatus of claim 9, wherein the input includes a plurality of cell images.

[0040] 11. The apparatus of claim 9, wherein the neural network includes two convolutional layers and a fully connected layer.

[0041] 12. The apparatus of claim 9, wherein the neural network includes multiple kernels for extracting features from the input with different levels of granularity.

[0042] 13. The apparatus of claim 9, wherein the noise events are selected from the group consisting of clumps, debris, doublets, edges, defocus, separation blebs, and blebs.

[0043] 14. The apparatus of claim 9, wherein classifying the input includes labeling each image as the noise event or the single cell.

[0044] 15. The apparatus of claim 9, wherein the neural network is configured to classify more than 2,000 images per second.

[0045] 16. The apparatus of claim 9, wherein the application is further configured to implement a bright field model or a fluorescence model in parallel with classifying the input.

[0046] 17. A system comprising: a first device configured to acquire a cell image; receiving the cell image; processing the cell images using a neural network to identify noise events or single cells; classifying the cell image as the noise event or the single cell; a second device configured to perform the A system including:

[0047] 18. The system of claim 17, wherein the neural network includes two convolutional layers and a fully connected layer.

[0048] 19. The system described in paragraph 17, wherein the neural network includes multiple kernels for extracting features with different levels of granularity from the cell image.

[0049] 20. The system of claim 17, wherein the noise events are selected from the group consisting of clumps, debris, doublets, edges, defocus, separation blebs, and blebs.

[0050] 21. The system described in paragraph 17, wherein classifying the cell image includes labeling the noise event or the single cell in the cell image.

[0051] 22. The system of claim 17, wherein the neural network is configured to classify more than 2,000 images per second.

[0052] 23. The system of claim 17, wherein the second device is further configured to implement a bright field model or a fluorescence model in parallel with classifying the cell images.

[0053] The present invention has been described with reference to specific embodiments incorporating details to facilitate an understanding of the principles of construction and operation of the invention. Reference herein to specific embodiments and their details is not intended to limit the scope of the claims appended hereto. Those skilled in the art will readily appreciate that various other modifications can be made to the embodiments chosen for illustration without departing from the spirit and scope of the invention as defined by the claims. [Explanation of symbols]

[0054] 100 aggregates 102 Doublet 104 Debris 106 Edge 108 Out of Focus 110 Separated Bleb 112 Bleb 114 single cells 200 Neural Networks 300 Receive input on a device 302 Process input using neural networks 304 Classify Input 400 Computer equipment 402 Network Interface 404 Memory 406 processor 408 I / O Devices 410 Bus 412 Storage device 420 Single Cell Identification Hardware 430 Single Cell Identification Applications

Claims

1. A method programmed into a non-transitory memory of a device, comprising: receiving an input; processing the input using a neural network to identify noise events or single cells; classifying the input as the noise event or the single cell; A method comprising:

2. The method of claim 1 , wherein the input comprises a plurality of cell images.

3. The method of claim 1 , wherein the neural network includes two convolutional layers and a fully connected layer.

4. 10. The method of claim 1, wherein the neural network includes multiple kernels for extracting features from the input with different fine-grained levels.

5. 2. The method of claim 1, wherein the noise event is selected from the group consisting of a cluster, debris, doublet, edge, defocus, detached-bleb, and bleb.

6. The method of claim 1 , wherein classifying the input comprises labeling each image as the noise event or the single cell.

7. 10. The method of claim 1, wherein the neural network is configured to classify more than 2,000 images per second.

8. 10. The method of claim 1, further comprising implementing a bright field model or a fluorescence model in parallel with the step of classifying the input.

9. 1. An apparatus comprising: A non-transitory memory for storing an application, said application comprising: Receiving input; processing the input using a neural network to identify noise events or single cells; classifying the input as the noise event or the single cell; a non-transitory memory for performing a processor coupled to the memory and configured to process the application; 10. An apparatus comprising:

10. The apparatus of claim 9 , wherein the input comprises a plurality of cell images.

11. The apparatus of claim 9 , wherein the neural network includes two convolutional layers and a fully connected layer.

12. 10. The apparatus of claim 9, wherein the neural network includes multiple kernels for extracting features with different levels of granularity from the input.

13. 10. The apparatus of claim 9, wherein the noise events are selected from the group consisting of clumps, debris, doublets, edges, defocus, separation blebs, and blebs.

14. 10. The apparatus of claim 9, wherein classifying the input comprises labeling each image as the noise event or the single cell.

15. 10. The apparatus of claim 9, wherein the neural network is configured to classify more than 2,000 images per second.

16. 10. The apparatus of claim 9, wherein the application is further configured to implement a bright field model or a fluorescence model in parallel with classifying the input.

17. 1. A system comprising: a first device configured to acquire a cell image; receiving the cell image; processing the cell images using a neural network to identify noise events or single cells; classifying the cell image as the noise event or the single cell; a second device configured to perform the A system comprising:

18. 18. The system of claim 17, wherein the neural network includes two convolutional layers and a fully connected layer.

19. 18. The system of claim 17, wherein the neural network includes multiple kernels for extracting features with different levels of granularity from the cell images.

20. 20. The system of claim 17, wherein the noise event is selected from the group consisting of a cluster, debris, doublet, edge, defocus, separation bleb, and bleb.

21. 18. The system of claim 17, wherein classifying the cell image comprises labeling the cell image as the noise event or the single cell.

22. 20. The system of claim 17, wherein the neural network is configured to classify more than 2,000 images per second.

23. 20. The system of claim 17, wherein the second device is further configured to implement a bright field model or a fluorescence model in parallel with classifying the cell images.

Images