Cell function annotation method, device, equipment and medium

Abstract

The present application discloses a cell function annotation method, device, equipment and medium, which relates to the field of artificial intelligence. The method includes: obtaining sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell; generating a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial positions in the tissue sample; based on the gene sequencing matrix, performing cell function annotation by deep learning to obtain a cell function annotation result. The scheme provided by the embodiment of the present application can improve the accuracy of cell function annotation.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of artificial intelligence, and in particular to a cell function annotation method, device, equipment and medium. Background Art

[0002] The process of analyzing the transcriptome sequencing data of cells, understanding the selective gene expression of cells at the molecular level, and then classifying cells at the molecular level is called cell function annotation.

[0003] In related technologies, before performing cell function annotation, cells are first extracted at multiple spatial sites using high-magnification imaging technology of tissue samples (such as tissue section imaging, confocal fluorescence imaging, etc.), and single-cell transcriptome sequencing is performed separately to obtain single-cell spatial transcriptome data; then, the single-cell spatial transcriptome data is analyzed using a clustering method to obtain cell function annotation.

[0004] However, since cell function is determined by the selective expression of cell genes and the interactions between multiple cells in tissues, the accuracy of cell function annotation using clustering is poor. Summary of the invention

[0005] The present application embodiment provides a cell function annotation method, device, equipment and medium, which can improve the accuracy of cell function annotation. The technical solution is as follows:

[0006] On the one hand, the present application embodiment provides a cell function annotation method, the method comprising:

[0007] Acquire sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell;

[0008] Generating a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial locations in the tissue sample;

[0009] Based on the gene sequencing matrix, cell function annotation is performed by deep learning to obtain cell function annotation results.

[0010] On the other hand, an embodiment of the present application provides a cell function annotation device, the device comprising:

[0011] A first acquisition module is used to acquire sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell;

[0012] A first generating module, configured to generate a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial locations in the tissue sample;

[0013] The first annotation module is used to perform cell function annotation based on the gene sequencing matrix through deep learning to obtain cell function annotation results.

[0014] On the other hand, an embodiment of the present application provides a computer device, which includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the cell function annotation method described in the above aspects.

[0015] On the other hand, an embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the cell function annotation method as described in the above aspects.

[0016] On the other hand, an embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the cell function annotation method provided by the above aspect.

[0017] In an embodiment of the present application, after the sequencing data of the tissue sample is obtained, a gene sequencing matrix of the tissue sample is generated based on the sequencing data. Since the gene sequencing matrix can reflect the gene distribution at different spatial locations in the tissue sample, that is, the gene sequencing matrix integrates the data of the two modalities of gene expression and spatial position, and deep learning has advantages in processing structured data, therefore, through deep learning, when performing cell function annotation based on the gene sequencing matrix, the global characteristics of the tissue sample and the local characteristics of the local cell population can be comprehensively considered, which helps to improve the accuracy of cell function annotation. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0019] Figure 1 is a schematic diagram of the principle of the cell function annotation process shown in an exemplary embodiment of the present application;

[0020] Figure 2 is a schematic diagram of an implementation environment provided by an exemplary embodiment of the present application;

[0021] Figure 3 is a flow chart of a cell function annotation method provided by an exemplary embodiment of the present application;

[0022] Figure 4 is a schematic diagram of capture points and cells in an in situ capture technology shown in an exemplary embodiment of the present application;

[0023] Figure 5 is a schematic diagram of different functional areas in a tissue sample shown in an exemplary embodiment of the present application;

[0024] Figure 6 is a flow chart of a cell function annotation method provided by another exemplary embodiment of the present application;

[0025] Figure 7 It is a schematic diagram of an implementation of a gene sequencing matrix generation and dimensionality reduction process shown in an exemplary embodiment of the present application;

[0026] Figure 8 is a structural diagram of a function partitioning model shown in an exemplary embodiment of the present application;

[0027] Fig. 9 is a flow chart of a function segmentation model training process shown in an exemplary embodiment of the present application;

[0028] Fig.10 is a schematic diagram of an implementation of a mask generation process shown in an exemplary embodiment of the present application;

[0029] Fig.11 is a schematic diagram of the principle of the cell function annotation process shown in another exemplary embodiment of the present application;

[0030] Fig.12 is a structural block diagram of a cell function annotation device provided by an exemplary embodiment of the present application;

[0031] Fig.13 A schematic diagram of the structure of a computer device provided by an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION

[0032] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0033] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0034] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0035] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0036] The solution provided in the embodiment of the present application, namely the application of machine learning / deep learning in the medical field, is used to perform cell function annotation on the gene sequencing matrix of the tissue sample (generated based on the sequencing data of the tissue sample) through a deep learning method to determine the cell function of each cell in the tissue sample.

[0037] like Figure 1As shown, in the cell function annotation scheme provided by the embodiment of the present application, after sequencing the tissue sample 11 by a sequencing instrument, sequencing data including gene expression data 12 and spatial position data 13 are obtained. The computer device fuses the data of these two modalities of gene expression data 12 and spatial position data 13 (a series of data preprocessing needs to be performed during the fusion process), and obtains the gene sequencing matrix 14 corresponding to the tissue sample 11. Further, using the advantage of deep learning when processing structured data, the computer device inputs the gene sequencing matrix 14 into the function segmentation model 15 obtained by pre-training, and the function segmentation model performs functional area segmentation to obtain the cell function annotation result 16 corresponding to the cell (or cell group) in the tissue sample 11. Because the function segmentation model performs functional area segmentation, the global characteristics of the tissue sample and the local characteristics of the local cell group can be comprehensively considered, therefore, compared to the clustering algorithm, the scheme provided by the embodiment of the present application can improve the accuracy of cell function annotation.

[0038] Figure 2 A schematic diagram of an implementation environment provided by an exemplary embodiment of the present application is shown. The implementation environment includes a terminal 210 and a server 220. The terminal 210 and the server 220 perform data communication via a communication network. Optionally, the communication network may be a wired network or a wireless network, and the communication network may be at least one of a local area network, a metropolitan area network, and a wide area network.

[0039] The terminal 210 is an electronic device that has a requirement for cell function annotation, and the electronic device may be a smart phone, a tablet computer, or a personal computer, etc., which is not limited in this embodiment. Figure 2 In the description, the terminal 210 is taken as an example of a personal computer used by scientific researchers.

[0040] The server 220 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms. In this embodiment, the server 220 is a server for providing cell function annotation services.

[0041] In some embodiments, Figure 2As shown, when it is necessary to perform functional annotation on cells in a tissue sample, or when it is necessary to perform functional regional division on a tissue sample, the scientific researcher will sequence the tissue sample through a sequencing device to obtain sequencing data, and import the sequencing data into a terminal 210. The terminal 210 uploads the sequencing data to a server 220, which performs data preprocessing and data fusion on the sequencing data to obtain a gene sequencing matrix. The server 220 inputs the gene sequencing matrix into a pre-trained functional segmentation model, and performs functional segmentation by the functional segmentation model to obtain a cell function annotation result. The server 220 feeds back the cell function annotation result to the terminal 210 for display by the terminal 210.

[0042] In other possible implementations, the terminal 210 may also pre-process the sequencing data and generate a gene sequencing matrix, thereby reporting the gene sequencing matrix to the server 220, and the server 220 may perform cell function annotation, or the terminal 210 may independently perform cell function annotation, which is not limited in this embodiment. For the convenience of description, the following embodiments are described by taking the cell function annotation method executed by a computer device as an example.

[0043] Figure 3 A flowchart of a cell function annotation method provided by an exemplary embodiment of the present application is shown. This embodiment is described by taking the method used in a computer device as an example, and the method includes the following steps.

[0044] Step 301, obtaining sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell.

[0045] The tissue sample may be a pathological slice such as a brain tissue slice or a tumor tissue slice. The embodiment of the present application does not limit the specific type of the tissue sample.

[0046] The sequencing data is obtained by sequencing tissue samples using a gene sequencing device. Optionally, the sequencing data is collected using a spatial transcriptome sequencing technology, including but not limited to: seqFISH, MERFISH, seqFISH+, osmFISH, Slide-seq, Visium, STARmap, HDST. The present application embodiment does not limit the specific type of sequencing data.

[0047] In some embodiments, the sequencing data includes gene expression data and spatial position data corresponding to each sequencing point, wherein the gene expression data is used to characterize the gene corresponding to the transcript sequenced at the sequencing point, and the spatial position data is used to characterize the spatial position of the sequencing point in the tissue sample.

[0048] In one possible implementation, a computer device determines the gene corresponding to the transcript (belonging to a pre-set gene library, and different transcripts may correspond to the same gene) based on the transcript sequenced at the sequencing point, thereby obtaining the gene expression data corresponding to the sequencing point. The gene expression data can be represented by a one-dimensional vector with a length of N, where N is the number of genes in the gene library, and each position value in the one-dimensional vector represents the number of genes.

[0049] In some embodiments, the sequencing data is collected using an in situ capture technique. When the resolution of the in situ capture technique is high, the sequencing point corresponds to a single cell (i.e., sequencing is performed on a single cell, and a single cell may cover multiple capture points captured in situ); when the resolution of the in situ capture technique is low, the sequencing point corresponds to a plurality of cells (bulk data). Figure 4 As shown, under the in situ capture technology, the spacing between the capture points 41 is 0.5 um. When a single cell 42 is sequenced, since the average diameter of the cell is between 10-20 um, the cell 42 covers more than 10×10 capture points 41.

[0050] Step 302: Generate a gene sequencing matrix of the tissue sample based on the sequencing data. The gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial locations in the tissue sample.

[0051] Since cell functions are determined by the selective expression of cell genes and the interactions between multiple cells in tissues, in order to improve the accuracy of cell function annotations, computer equipment fuses the two modalities of gene expression data and spatial position data in the sequencing data to construct a gene sequencing matrix that characterizes the distribution of genes at different spatial locations in tissue samples, so that functional annotations can be performed subsequently based on the gene sequencing matrix.

[0052] In some embodiments, the gene sequencing matrix is ​​a three-dimensional matrix with a size of H×W×C, wherein H and W can be measured in units of sequencing points (usually in the case of low resolution, characterizing the spatial position of the sequencing points in the tissue sample), or in units of cells (usually in the case of high resolution, characterizing the position of the cells in the tissue sample); C is the number of channels, and each gene corresponds to its own channel.

[0053] Optionally, before generating a gene sequencing matrix based on the sequencing data, the computer device also needs to perform data preprocessing on the sequencing data to eliminate the noise introduced during the sequencing data collection process and improve the accuracy of subsequent functional annotations, which will be described in detail in the following embodiments.

[0054] Step 303: Based on the gene sequencing matrix, cell function annotation is performed by deep learning to obtain cell function annotation results.

[0055] Since deep learning has better performance when processing structured data, in an embodiment of the present application, a computer device uses deep learning to perform cell function annotation on tissue samples based on a gene sequencing matrix to obtain cell function annotation results.

[0056] Optionally, the cell function annotation result corresponds to a sequencing point. When the sequencing point corresponds to a single cell, the cell function annotation result corresponding to the sequencing point is used to characterize the cell function of the single cell; when the sequencing point corresponds to multiple cells, the cell function annotation result corresponding to the sequencing point is used to characterize the cell function of the cell group.

[0057] In one possible implementation, a computer device pre-trains a deep learning model based on training samples through deep learning, and then uses the deep learning model to perform cell function annotation based on a structured gene sequencing matrix.

[0058] In some embodiments, the computer device can divide the tissue sample into functional areas according to the cell function annotation results, and visualize the divided functional areas. Figure 5 Different grayscale areas represent different functional areas in the tissue sample.

[0059] To summarize, in the embodiments of the present application, after the sequencing data of the tissue sample is obtained, a gene sequencing matrix of the tissue sample is generated based on the sequencing data. Since the gene sequencing matrix can reflect the gene distribution at different spatial locations in the tissue sample, that is, the gene sequencing matrix integrates the data of the two modalities of gene expression and spatial position, and deep learning has advantages in processing structured data, therefore, through deep learning, when performing cell function annotation based on the gene sequencing matrix, the global characteristics of the tissue sample and the local characteristics of the local cell population can be comprehensively considered, which helps to improve the accuracy of cell function annotation.

[0060] Since noise may be introduced during the sequencing data collection process, and the noise in the sequencing data will affect the accuracy of subsequent functional annotations, before generating a gene sequencing matrix based on the sequencing data, the computer equipment needs to perform quality control (i.e., data preprocessing) on ​​the sequencing data to eliminate the noise introduced during the data collection process. In addition, since different genes have different effects on cell functions, in order to further reduce the amount of calculation during subsequent cell function annotations, the computer equipment needs to perform gene selection to reduce the dimension of the gene sequencing matrix. The above process is described below using an exemplary embodiment.

[0061] Figure 6 A flowchart of a cell function annotation method provided by another exemplary embodiment of the present application is shown. This embodiment is described by taking the method used in a computer device as an example, and the method includes the following steps.

[0062] Step 601, obtaining sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell.

[0063] The implementation of this step may refer to the above step 301, and this embodiment will not be described in detail here.

[0064] Step 602: align the gene expression data and the spatial position data.

[0065] In one possible implementation, a computer device performs data alignment on gene expression data and spatial position data to determine the gene expression levels of various genes in the cell, thereby constructing a gene expression vector corresponding to the cell (or cell group), which can be a one-dimensional vector with a length of N, where N is the total number of genes in the gene library.

[0066] For example, when the gene library (preset) contains 10,000 genes, the gene expression vector of the cell is a one-dimensional vector with a length of 10,000. When the first bit in the one-dimensional vector is 10, it means that the gene expression level of the first gene in the cell is 10, and when the second bit in the one-dimensional vector is 0, it means that the cell does not contain the second gene.

[0067] Step 603: Filter out abnormal sequencing points among the sequencing points based on the gene expression data and spatial position data after data alignment.

[0068] The number of transcripts in a single cell or the gene expression of the gene corresponding to the transcript is usually within a certain range. When multiple cells overlap, cell death or premature cell rupture causes premature diffusion of mRNA, the number of transcripts or gene expression of a single cell will exceed the range, thereby affecting the accuracy of subsequent functional annotations. Therefore, after the computer device completes the data alignment, it is necessary to identify abnormalities in the sequencing points, thereby filtering out abnormal sequencing points. In one possible embodiment, the computer device filtering abnormal sequencing points may include the following steps:

[0069] 1. Based on the gene expression data and spatial location data after data alignment, determine the number of transcripts and gene expression levels of each cell.

[0070] Optionally, the computer device counts the total amount of transcripts in the cell and the total amount of genes corresponding to the transcripts based on the aligned data to obtain the number of transcripts and the gene expression level.

[0071] 2. Determine cells whose transcript quantity does not fall within the quantity interval or whose gene expression level does not fall within the expression level interval as abnormal cells.

[0072] In a possible embodiment, the computer device is provided with a quantity region of transcripts in normal cells, and an expression range of genes in normal cells. When the number of transcripts of a cell is within the quantity range, and the gene expression is within the expression range, the computer determines that the cell is a normal cell; when the number of transcripts of a cell is outside the quantity range, or the gene expression is outside the expression range, the computer determines that the cell is an abnormal cell. Among them, the quantity region and the expression range can be pre-set by the R&D personnel. For example, the quantity range is 2000-10000, and the expression range is 50-200, and this embodiment does not constitute a limitation on this.

[0073] 3. Filter abnormal sequencing points corresponding to abnormal cells.

[0074] Furthermore, the computer device determines the sequencing points corresponding to the abnormal cells as abnormal sequencing points, and filters the abnormal sequencing points. The gene sequencing matrix generated subsequently does not include the sequencing data corresponding to the abnormal sequencing points.

[0075] Step 604: Generate a gene distribution matrix corresponding to the gene based on the gene expression data and spatial position data after data alignment. The gene distribution matrix is ​​used to represent the gene expression levels of the gene at different spatial positions.

[0076] After completing the preprocessing, the computer equipment determines all genes contained in the cells in the tissue sample, and determines the gene expression of various genes at different spatial locations in the tissue sample, and generates a gene distribution matrix (two-dimensional matrix) corresponding to each gene. The size of the gene distribution matrix is ​​H×W.

[0077] In a possible implementation, when cells are used as units, the computer device removes duplicate genes contained in each cell in the tissue sample, determines the total amount of gene types contained in the tissue sample, and generates a gene distribution matrix based on the number of the same gene in different cells. For example, the gene distribution matrix corresponding to gene 1 is used to represent the gene expression amount of gene 1 in each cell in the tissue sample.

[0078] Indicatively, Figure 7 As shown, the computer device determines that the total number of gene types in the tissue sample is C, and generates a corresponding gene distribution matrix for each gene. Different matrix positions in the gene distribution matrix correspond to different spatial positions, and a total of C gene distribution matrices are generated.

[0079] Step 605: Generate a gene sequencing matrix based on the gene distribution matrices corresponding to different genes. The gene sequencing matrix is ​​a three-dimensional matrix.

[0080] Furthermore, the computer equipment performs three-dimensional splicing on the gene distribution matrices corresponding to different genes to obtain a three-dimensional gene sequencing matrix, wherein the two-dimensional matrix corresponding to each channel in the gene sequencing matrix is ​​the gene distribution matrix corresponding to the gene.

[0081] Indicatively, Figure 7 As shown, the computer equipment performs three-dimensional splicing on C H×W gene distribution matrices to obtain an H×W×C gene sequencing matrix.

[0082] Optionally, in order to eliminate the differences in sequencing data collection at different spatial locations (which may be caused by uneven RNA separation, inconsistent sequencing depth, etc.), the computer equipment first normalizes the gene expression levels in each gene distribution matrix to obtain a normalized gene distribution matrix, and then performs three-dimensional splicing on the normalized gene distribution matrix to obtain a gene sequencing matrix.

[0083] Step 606, based on the gene sequencing matrix, determine the highly expressed and highly variable genes.

[0084] The dimension of the gene sequencing matrix constructed by the above steps is usually high, and the amount of computation required to perform functional annotation directly based on the gene sequencing matrix is ​​large. In addition, the gene sequencing matrix may contain a large number of genes that have little impact on functional annotation, and filtering the sequencing data corresponding to these genes will not affect the accuracy of functional annotation. Therefore, under the premise of ensuring the accuracy of functional annotation, in order to reduce the amount of computation, the computer equipment further screens genes and reduces the dimension of the gene sequencing matrix.

[0085] Since cell functions are greatly affected by highly expressed genes and highly variable genes, in the embodiment of the present application, the computer device screens out highly expressed and highly variable genes based on the gene sequencing matrix. Optionally, the process may include the following steps:

[0086] 1. Based on the gene sequencing matrix, the average expression intensity and variability variance of each gene are determined. The average expression intensity is determined based on the total gene expression and the number of sequencing points. The variability variance is used to characterize the difference in gene expression of the same gene at different sequencing points.

[0087] In a possible implementation, for each channel in the gene sequencing matrix, the computer device sums the values ​​of each matrix position in the gene distribution matrix corresponding to the channel to obtain the total gene expression of the gene, thereby determining the average expression intensity of the gene based on the total gene expression and the number of sequencing points. Further, the computer device determines the variability variance corresponding to the gene based on the gene expression and the average expression intensity of the gene at each sequencing point, wherein the larger the variability variance, the greater the difference in gene expression of the same gene at different sequencing points.

[0088] 2. Identify highly expressed and highly variable genes based on average expression intensity and variability variance.

[0089] In a possible implementation, the computer device determines genes whose average expression intensity is higher than an intensity threshold and whose variability variance is greater than a variance threshold as highly expressed and highly variable genes. The intensity threshold and the variance threshold can be customized by the R&D personnel.

[0090] In another possible embodiment, the computer device determines the genes with the top k1 average expression intensities as highly expressed genes, and then determines the genes with the top k1 variability variances among the highly expressed genes as highly expressed and highly variable genes.

[0091] Of course, the computer device can use other methods such as chart analysis to determine highly expressed and highly variable genes, and the embodiments of the present application are not limited to this.

[0092] Step 607, performing dimensionality reduction processing on the gene sequencing matrix based on the highly expressed and highly variable genes, wherein the gene sequencing matrix after the dimensionality reduction processing includes a gene distribution matrix corresponding to the highly expressed and highly variable genes.

[0093] After determining the highly expressed and highly variable genes, the computer device deletes the gene distribution matrix corresponding to other genes other than the highly expressed and highly variable genes in the gene sequencing matrix, thereby obtaining the gene sequencing matrix after dimensionality reduction processing.

[0094] Indicatively, Figure 7 As shown, the computer device performs dimensionality reduction processing on the gene sequencing matrix according to the determined C' highly expressed and highly variable genes to obtain a gene sequencing matrix of H×W×C'.

[0095] Step 608, input the gene sequencing matrix into the functional segmentation model to obtain the functional segmentation results output by the functional segmentation model. The functional segmentation model is trained by deep learning, and the functional segmentation results include the cell function annotation results corresponding to the sequencing points.

[0096] In a possible implementation, a computer device is provided with a functional segmentation model trained by deep learning. The computer device uses the gene sequencing matrix as the input of the functional segmentation model, annotates the cells corresponding to each sequencing point through the functional segmentation model, and outputs the cell function annotation results. The gene sequencing matrix can be regarded as a multi-channel image, and the functional segmentation model is used to perform pixel-level annotation on the multi-channel image.

[0097] Optionally, the function segmentation model includes an encoder consisting of multiple convolutional layers and a decoder consisting of multiple deconvolutional layers. Figure 8As shown, the convolution layer in the encoder 82 is used to perform a sliding window convolution operation on the input gene sequencing matrix 81, and then the pooling layer is used to reduce the dimension. Finally, the encoder 82 outputs a feature map with a dimension lower than the gene sequencing matrix 81. This layered calculation operation of the encoder 82 obtains a nonlinear, local-to-global feature map that combines the local perceived saliency features and the global style. The deconvolution layer in the decoder 83 performs a sliding window deconvolution operation on the feature map output by the encoder 82, and gradually restores the feature map size to the original input size. In this process, some feature maps of the convolution layer can be introduced from the encoder 82 for fusion to improve the recognition resolution of the decoder 83 (i.e., the U-net structure). The function segmentation model uses a softmax multi-classifier to normalize the output of the decoder 83 to obtain a function segmentation result 84.

[0098] It should be noted that in addition to the above-mentioned full-convolutional network, the function segmentation model can also adopt other network structures such as EfficientNet, ResNet and DenseNet. The embodiments of the present application do not limit the specific network structure of the function segmentation model.

[0099] In this embodiment, before the computer device generates a gene sequencing matrix based on the sequencing data, it filters abnormal sequencing points based on the number of transcripts and gene expression levels of the cells to avoid the sequencing data of the abnormal sequencing points from affecting subsequent functional annotations, which helps to improve the accuracy of the functional annotations.

[0100] In addition, computer equipment can reduce the dimensionality of gene sequencing matrices by identifying highly expressed and highly variable genes, thus avoiding the waste of computing resources caused by subsequent operations on data corresponding to low-expressed and low-variable genes, and helping to increase the speed of functional annotation.

[0101] In a possible implementation, Fig. 9 As shown, the training process of the functional segmentation model may include the following steps.

[0102] Step 901, based on the standard functional partition map, generate a functional segmentation mask corresponding to the training sample, the standard sample corresponding to the standard functional partition map and the training sample belong to the same tissue type, and the functional segmentation mask contains the cell function annotations corresponding to each sample sequencing point in the training sample.

[0103] Before model training, the computer device first generates a functional segmentation mask based on the standard functional partition map corresponding to the standard sample of the same tissue type as the training sample. The functional segmentation mask is the ground truth for model training. For example, when the training sample is a brain tissue sample, the standard functional partition map is used to indicate the division of each functional partition in the standard brain tissue sample.

[0104] Since the standard sample is not completely consistent with the training sample, that is, there are differences in the tissue contours of the standard sample and the training sample, the computer equipment needs to perform image space alignment on the standard functional partition map and the training sample so that the contour of the standard functional partition map matches the contour of the training sample, thereby generating a functional segmentation mask corresponding to the training sample based on the standard functional partition map after image space alignment.

[0105] Optionally, the image space alignment method may include rotation, scaling, affine transformation, etc., which is not limited in this embodiment.

[0106] Optionally, the computer device assigns functional labels to sequencing points in the functional partitions in the training samples according to the mapping relationship between the standard functional partitions and the training sample images, thereby obtaining a functional segmentation mask.

[0107] Indicatively, Fig.10 As shown, the computer device performs image space alignment on the standard functional partition map 1001 and the training sample 1002, and performs functional partition mapping after the alignment to obtain a functional segmentation mask 1003.

[0108] Step 902: Obtain sample sequencing data of training samples.

[0109] The sample sequencing data obtained during the training process is similar to the test data obtained during the actual application process, and will not be described in detail in this embodiment.

[0110] Step 903: Generate a sample gene sequencing matrix of the training sample based on the sample sequencing data.

[0111] The process of generating the sample gene sequencing matrix during the training process can refer to the above application process, and this embodiment will not be described in detail here.

[0112] Step 904: input the sample gene sequencing matrix into the functional segmentation model to obtain the functional segmentation prediction result output by the functional segmentation model.

[0113] The implementation process of this step can refer to the above step 608, and this implementation will not be described in detail here.

[0114] Step 905 , using the function segmentation mask as supervision of the function segmentation prediction result to train the function segmentation model.

[0115] In a possible implementation, the computer device calculates the segmentation loss between the function segmentation prediction result and the function segmentation mask, and then based on the segmentation loss, uses the back propagation algorithm to update the model parameters of the function segmentation model. When the training end condition is met (reaching the number of iterations, or parameter convergence), the computer device stops model training. The segmentation loss can be a cross entropy loss.

[0116] In order to further improve the accuracy of functional annotation, in a possible implementation, the computer device can also obtain the cell morphology features and cell texture features of the cells in the tissue sample, and perform feature fusion on the cell morphology features, cell texture features and gene sequencing matrix to obtain a fusion feature matrix, thereby inputting the fusion feature matrix into the functional segmentation model to obtain the functional segmentation results output by the functional segmentation model. When performing feature fusion, the computer device needs to align the cell texture features and cell morphology features with each sequencing point on the gene sequencing matrix.

[0117] In some embodiments, the cell morphological features are obtained by extracting features from the cell contours. For example, the cell morphological features are extracted by image processing tools such as openCV, including but not limited to: cell area, cell perimeter, cell contour complexity, cell equivalent diameter, cell ellipticity, cell polarity, etc.

[0118] In some embodiments, the cell texture feature is determined based on the pixel values ​​of pixel points within the cell range in the tissue sample image, wherein the tissue sample image can use various staining methods and various original resolutions, including but not limited to: H&E staining, ssDNA staining, DAPI staining, etc., and the original resolution can be 10x, 20x, 40x, etc. of the microscope.

[0119] Optionally, the cell texture feature can be calculated using a local binary pattern (LBP) algorithm. For example, the computer device compares the pixel values ​​of a pixel with those of the surrounding eight pixels to obtain the 8-bit binary code corresponding to the pixel, and converts the binary code into an integer of 0-255, thereby using the integer as the LBP value of the pixel in the LBP map. Finally, the computer device performs a histogram conversion on the LBP values ​​of all pixels within the cell range to obtain the cell's texture feature vector.

[0120] In an illustrative example, Fig.11As shown, after sequencing the tissue sample 11 by the sequencing instrument, sequencing data including gene expression data 12 and spatial position data 13 are obtained. The computer device obtains the gene sequencing matrix 14 corresponding to the tissue sample 11 by fusing the data of the two modalities of gene expression data 12 and spatial position data 13 (a series of data preprocessing is required during the fusion process). At the same time, the computer device extracts morphological features from the image of the tissue sample 11 to obtain cell morphological features 17, and extracts texture features from the image of the tissue sample 11 to obtain cell texture features 18. Further, the computer device fuses the gene sequencing matrix 14, cell morphological features 17 and cell texture features 18 to obtain a fused feature matrix 19. The computer device inputs the fused feature matrix 19 into the pre-trained functional segmentation model 15, and the functional segmentation model performs functional area segmentation to obtain the cell function annotation result 16 corresponding to the cells (or cell groups) in the tissue sample 11.

[0121] It should be noted that when the input of the functional segmentation model is a fused feature matrix, during the model training process, it is also necessary to extract morphological and texture features of the training samples, and to fuse the sample cell morphological features, sample cell texture features and sample gene sequencing matrix. This embodiment will not go into details.

[0122] Fig.12 : is a structural block diagram of a cell function annotation device provided by an exemplary embodiment of the present application, the device comprising:

[0123] A first acquisition module 1201 is used to acquire sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell;

[0124] A first generating module 1202 is used to generate a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial positions in the tissue sample;

[0125] The first annotation module 1203 is used to perform cell function annotation based on the gene sequencing matrix through deep learning to obtain cell function annotation results.

[0126] Optionally, the first generating module 1202 includes:

[0127] A data alignment unit, used for aligning the gene expression data and the spatial position data;

[0128] A first generating unit is used to generate a gene distribution matrix corresponding to the gene based on the gene expression data and the spatial position data after data alignment, wherein the gene distribution matrix is ​​used to characterize the gene expression amount of the gene at different spatial positions;

[0129] The second generating unit is used to generate the gene sequencing matrix based on the gene distribution matrix corresponding to different genes, and the gene sequencing matrix is ​​a three-dimensional matrix.

[0130] Optionally, the first generating module 1202 further includes:

[0131] A filtering unit is used to filter abnormal sequencing points among the sequencing points based on the gene expression data and the spatial position data after data alignment.

[0132] Optionally, the filtering unit is used to:

[0133] Determining the number of transcripts and the gene expression level of each cell based on the gene expression data and the spatial position data after data alignment;

[0134] Determining cells whose transcript quantity does not fall within the quantity interval, or whose gene expression level does not fall within the expression level interval, as abnormal cells;

[0135] The abnormal sequencing points corresponding to the abnormal cells are filtered.

[0136] Optionally, the second generating unit is used to:

[0137] Normalizing the gene expression levels in each of the gene distribution matrices to obtain a normalized gene distribution matrix;

[0138] The normalized gene distribution matrix is ​​three-dimensionally spliced ​​to obtain the gene sequencing matrix.

[0139] Optionally, the device further comprises:

[0140] A determination module, used for determining highly expressed and highly variable genes based on the gene sequencing matrix;

[0141] A dimensionality reduction module is used to perform dimensionality reduction processing on the gene sequencing matrix based on the highly expressed and highly variable genes, wherein the gene sequencing matrix after dimensionality reduction processing contains the gene distribution matrix corresponding to the highly expressed and highly variable genes.

[0142] Optionally, the determining module includes:

[0143] A first determination unit is used to determine the average expression intensity and variability variance of each gene based on the gene sequencing matrix, wherein the average expression intensity is determined based on the total expression amount of the gene and the number of sequencing points, and the variability variance is used to characterize the difference in gene expression amount of the same gene at different sequencing points;

[0144] A second determination unit is used to determine the highly expressed and highly variable genes based on the average expression intensity and the variability variance.

[0145] Optionally, the first annotation module 1203 is used to:

[0146] The gene sequencing matrix is ​​input into a functional segmentation model to obtain a functional segmentation result output by the functional segmentation model, wherein the functional segmentation model is trained by deep learning, and the functional segmentation result includes the cell function annotation result corresponding to the sequencing point.

[0147] Optionally, the device further comprises:

[0148] A mask generation module, used to generate a functional segmentation mask corresponding to a training sample based on a standard functional partition map, wherein the standard sample corresponding to the standard functional partition map and the training sample belong to the same tissue type, and the functional segmentation mask includes a cell function annotation corresponding to each sample sequencing point in the training sample;

[0149] A second acquisition module, used to acquire sample sequencing data of the training sample;

[0150] A second generating module, used to generate a sample gene sequencing matrix of the training sample based on the sample sequencing data;

[0151] A second annotation module is used to input the sample gene sequencing matrix into the functional segmentation model to obtain the functional segmentation prediction result output by the functional segmentation model;

[0152] A training module is used to train the function segmentation model by using the function segmentation mask as supervision of the function segmentation prediction result.

[0153] Optionally, the mask generating module is used to:

[0154] Performing image space alignment on the standard functional partition and the training sample;

[0155] Based on the standard functional partition after image space alignment, the functional segmentation mask corresponding to the training sample is generated.

[0156] Optionally, the device further comprises:

[0157] A third acquisition module is used to acquire cell morphological features and cell texture features of cells in the tissue sample, wherein the cell morphological features are obtained by extracting features of cell contours, and the cell texture features are determined based on pixel values ​​of pixel points within a cell range in the tissue sample image;

[0158] The first annotation module 1203 is further used for:

[0159] Performing feature fusion on the cell morphology feature, the cell texture feature and the gene sequencing matrix to obtain a fusion feature matrix;

[0160] The fusion feature matrix is ​​input into a functional segmentation model to obtain a functional segmentation result output by the functional segmentation model, wherein the functional segmentation model is trained by deep learning, and the functional segmentation result includes the cell function annotation result corresponding to the sequencing point.

[0161] To summarize, in the embodiments of the present application, after the sequencing data of the tissue sample is obtained, a gene sequencing matrix of the tissue sample is generated based on the sequencing data. Since the gene sequencing matrix can reflect the gene distribution at different spatial locations in the tissue sample, that is, the gene sequencing matrix integrates the data of the two modalities of gene expression and spatial position, and deep learning has advantages in processing structured data, therefore, through deep learning, when performing cell function annotation based on the gene sequencing matrix, the global characteristics of the tissue sample and the local characteristics of the local cell population can be comprehensively considered, which helps to improve the accuracy of cell function annotation.

[0162] It should be noted that: the device provided in the above embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the implementation process thereof is detailed in the method embodiment, which will not be repeated here.

[0163] Please refer to Fig.13, which shows a schematic diagram of the structure of a computer device provided by an exemplary embodiment of the present application. Specifically, the computer device 1300 includes a central processing unit (CPU) 1301, a system memory 1304 including a random access memory 1302 and a read-only memory 1303, and a system bus 1305 connecting the system memory 1304 and the central processing unit 1301. The computer device 1300 also includes a basic input / output system (I / O system) 1306 that helps transmit information between various devices in the computer, and a large-capacity storage device 1307 for storing an operating system 1313, application programs 1314 and other program modules 1315.

[0164] The basic input / output system 1306 includes a display 1308 for displaying information and an input device 1309 such as a mouse and a keyboard for user inputting information. The display 1308 and the input device 1309 are connected to the central processing unit 1301 through an input / output controller 1310 connected to the system bus 1305. The basic input / output system 1306 may also include an input / output controller 1310 for receiving and processing inputs from a plurality of other devices such as a keyboard, a mouse, or an electronic stylus. Similarly, the input / output controller 1310 also provides output to a display screen, a printer, or other types of output devices.

[0165] The mass storage device 1307 is connected to the central processing unit 1301 through a mass storage controller (not shown) connected to the system bus 1305. The mass storage device 1307 and its associated computer readable media provide non-volatile storage for the computer device 1300. That is, the mass storage device 1307 may include a computer readable medium (not shown) such as a hard disk or drive.

[0166] Without loss of generality, the computer-readable medium may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer storage media include random access memory (RAM), read-only memory (ROM), flash memory or other solid-state storage technology, compact disc read-only memory (CD-ROM), digital versatile disc (DVD) or other optical storage, cassette, magnetic tape, disk storage or other magnetic storage devices. Of course, those skilled in the art will know that the computer storage medium is not limited to the above. The above-mentioned system memory 1304 and mass storage device 1307 can be collectively referred to as memory.

[0167] The memory stores one or more programs, and the one or more programs are configured to be executed by one or more central processing units 1301. The one or more programs contain instructions for implementing the above-mentioned methods. The central processing unit 1301 executes the one or more programs to implement the methods provided by the above-mentioned various method embodiments.

[0168] According to various embodiments of the present application, the computer device 1300 can also be connected to a remote computer on the network through a network such as the Internet. That is, the computer device 1300 can be connected to the network 1312 through the network interface unit 1311 connected to the system bus 1305, or the network interface unit 1311 can be used to connect to other types of networks or remote computer systems (not shown).

[0169] The memory also includes one or more programs, which are stored in the memory and include steps executed by a computer device in the method provided in the embodiment of the present application.

[0170] An embodiment of the present application also provides a computer-readable storage medium, in which at least one instruction is stored. The at least one instruction is loaded and executed by a processor to implement the cell function annotation method described in any of the above embodiments.

[0171] Optionally, the computer readable storage medium may include: ROM, RAM, solid state drives (SSD) or optical disks, etc. Among them, RAM may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0172] The embodiment of the present application provides a computer program product or a computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the cell function annotation method described in the above embodiment.

[0173] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0174] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A cell function annotation method, characterized in that: The method comprises: Acquire sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell; Generating a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial locations in the tissue sample; The gene sequencing matrix is ​​input into a functional segmentation model to obtain a functional segmentation result output by the functional segmentation model, wherein the functional segmentation result includes a cell function annotation result corresponding to the sequencing point. The functional segmentation model generates a functional segmentation mask corresponding to a training sample based on a standard functional partitioning diagram, and the functional segmentation mask is used as a supervised training for a functional segmentation prediction result. The functional segmentation prediction result is output by the functional segmentation model based on the sample gene sequencing matrix of the input training sample. The standard sample corresponding to the standard functional partitioning diagram belongs to the same tissue type as the training sample, and the functional segmentation mask includes a cell function annotation corresponding to each sample sequencing point in the training sample.

2. The method according to claim 1, characterized in that: The step of generating a gene sequencing matrix of the tissue sample based on the sequencing data comprises: Performing data alignment on the gene expression data and the spatial position data; Based on the gene expression data and the spatial position data after data alignment, a gene distribution matrix corresponding to the gene is generated, wherein the gene distribution matrix is ​​used to characterize the gene expression amount of the gene at different spatial positions; The gene sequencing matrix is ​​generated based on the gene distribution matrix corresponding to different genes, and the gene sequencing matrix is ​​a three-dimensional matrix.

3. The method according to claim 2, characterized in that After aligning the gene expression data and the spatial position data, the method further includes: Based on the gene expression data and the spatial position data after data alignment, abnormal sequencing points among the sequencing points are filtered.

4. The method according to claim 3, characterized in that The filtering of abnormal sequencing points among the sequencing points based on the gene expression data and the spatial position data after data alignment includes: Determining the number of transcripts and the gene expression level of each cell based on the gene expression data and the spatial position data after data alignment; Determining cells whose transcript quantity does not fall within the quantity interval, or whose gene expression level does not fall within the expression level interval, as abnormal cells; The abnormal sequencing points corresponding to the abnormal cells are filtered.

5. The method according to claim 2, characterized in that: The step of generating the gene sequencing matrix based on the gene distribution matrix corresponding to different genes includes: Normalizing the gene expression levels in each of the gene distribution matrices to obtain a normalized gene distribution matrix; The normalized gene distribution matrix is ​​three-dimensionally spliced ​​to obtain the gene sequencing matrix.

6. The method according to claim 2, characterized in that After generating the gene sequencing matrix based on the gene distribution matrix corresponding to different genes, the method further includes: Based on the gene sequencing matrix, determining highly expressed and highly variable genes; The gene sequencing matrix is ​​subjected to dimensionality reduction processing based on the highly expressed and highly variable genes, wherein the gene sequencing matrix after dimensionality reduction processing contains the gene distribution matrix corresponding to the highly expressed and highly variable genes.

7. The method according to claim 6, characterized in that Determining highly expressed and highly variable genes based on the gene sequencing matrix includes: Based on the gene sequencing matrix, the average expression intensity and variability variance of each gene are determined, wherein the average expression intensity is determined based on the total gene expression amount and the number of sequencing points, and the variability variance is used to characterize the difference in gene expression amount of the same gene at different sequencing points; The highly expressed and highly variable genes are determined based on the average expression intensity and the variability variance.

8. The method according to claim 1, characterized in that The method comprises: Performing image space alignment on the standard functional partition and the training sample; Based on the standard functional partition after image space alignment, the functional segmentation mask corresponding to the training sample is generated.

9. The method according to any one of claims 1 to 8, characterized in that: The method further comprises: Acquiring cell morphological features and cell texture features of cells in the tissue sample, wherein the cell morphological features are obtained by extracting features of cell contours, and the cell texture features are determined based on pixel values ​​of pixel points within a cell range in the tissue sample image; The step of inputting the gene sequencing matrix into a functional segmentation model to obtain a functional segmentation result output by the functional segmentation model comprises: Performing feature fusion on the cell morphology feature, the cell texture feature and the gene sequencing matrix to obtain a fusion feature matrix; The fused feature matrix is ​​input into the functional segmentation model to obtain the functional segmentation result output by the functional segmentation model. The functional segmentation model generates the functional segmentation mask corresponding to the training sample based on the standard functional partition map, and uses the functional segmentation mask as the supervised training of the functional segmentation prediction result. The functional segmentation prediction result is output by the functional segmentation model based on the sample fusion feature matrix of the input training sample. The sample fusion feature matrix is ​​obtained by fusing the sample cell morphology features, sample cell texture features and the sample gene sequencing matrix of the training sample.

10. A cell function annotation device, characterized in that: The device comprises: A first acquisition module is used to acquire sequencing data of a tissue sample, wherein the sequencing data includes gene expression data and spatial position data of each sequencing point, and the sequencing point corresponds to at least one cell; A first generating module, configured to generate a gene sequencing matrix of the tissue sample based on the sequencing data, wherein the gene sequencing matrix is ​​used to characterize the distribution of genes at different spatial locations in the tissue sample; The first annotation module is used to input the gene sequencing matrix into the functional segmentation model to obtain the functional segmentation results output by the functional segmentation model, wherein the functional segmentation results include the cell function annotation results corresponding to the sequencing points, the functional segmentation model generates a functional segmentation mask corresponding to the training sample based on a standard functional partitioning diagram, and uses the functional segmentation mask as the supervised training of the functional segmentation prediction result, the functional segmentation prediction result is output by the functional segmentation model based on the sample gene sequencing matrix of the input training sample, the standard sample corresponding to the standard functional partitioning diagram belongs to the same tissue type as the training sample, and the functional segmentation mask includes the cell function annotations corresponding to each sample sequencing point in the training sample.

11. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, and the at least one instruction is loaded and executed by the processor to implement the cell function annotation method according to any one of claims 1 to 9.

12. A computer-readable storage medium, characterized in that: The readable storage medium stores at least one instruction, and the at least one instruction is loaded and executed by a processor to implement the cell function annotation method according to any one of claims 1 to 9.

13. A computer program product, characterized in that The computer program product may include computer instructions, which are stored in a computer-readable storage medium; a processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device implements the cell function annotation method as described in any one of claims 1 to 9.

Images