Cell type annotation method and apparatus, and device and storage medium

Abstract

Provided in the present application are a cell type annotation method and apparatus, and a device and a storage medium. The method comprises: clustering cell sequencing data to obtain a plurality of cell populations; analyzing a differential high-expression component list of each cell population, wherein the differential high-expression component list includes differential high-expression components (the components refer to genes or proteins) in the cell population which are sequenced according to specificity scores in descending order; and for each cell population, according to the sequence of preset marker components corresponding to cell types in the differential high-expression component list of the cell population, determining the probability of the cell population belonging to each cell type, and determining a corresponding cell type with the highest probability to be the cell type to which the cell population belongs, wherein the sequence of the marker components corresponding to the cell types in the differential high-expression component list of the cell population is positively correlated with the probability of the cell population belonging to the cell type. In the present solution, the effect of a marker component on a cell type to which a cell population belongs is correspondingly enhanced or reduced according to the magnitude of the specificity of the marker component in the cell population, thereby improving the accuracy of an annotation result.

Description

A method, device, equipment and storage medium for cell type annotation Technical Field

[0001] The present application relates to the field of cell sequencing technology, and in particular to a method, apparatus, device and storage medium for annotating cell types. Background Art

[0002] Cell type annotation is an important part of cell sequencing and the basis for other subsequent downstream analyses.

[0003] Currently, a widely accepted annotation method in this field is annotation based on the distribution of marker genes (or marker proteins) in the clustering results. Specifically, after clustering multiple cell clusters based on sequencing results, the distribution of marker genes (or marker proteins) corresponding to a specific cell type is determined based on whether the differentially expressed genes (or differentially expressed proteins) of the cell clusters contain marker genes (or marker proteins) corresponding to that specific cell type.

[0004] The drawback of this method is that it does not take into account the specific ranking of marker genes among differentially expressed genes, that is, it assumes that each differentially expressed gene has the same determining effect on cell type. However, the actual situation should be that the more specific the differentially expressed gene, the greater its impact on cell type. Therefore, this existing method has low accuracy.

[0005] Summary of the Invention

[0006] The present application provides a method, apparatus, device and storage medium for cell type annotation to improve the accuracy of cell type annotation.

[0007] The first aspect of the present application provides a method for annotating cell types, comprising:

[0008] Acquire cell sequencing data of multiple cells, and cluster multiple cell populations obtained based on the cell sequencing data;

[0009] Obtaining a list of differentially highly expressed components of each cell population according to the cell sequencing data analysis; wherein the list of differentially highly expressed components comprises a plurality of differentially highly expressed components in the cell population sorted from high to low according to specificity scores, and the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins;

[0010] For each of the cell populations, the probability value of the cell population belonging to each of the cell types is determined based on the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and the cell type to which the cell population belongs is determined based on the probability value; wherein the marker component is a marker gene or a marker protein.

[0011] Optionally, also include:

[0012] The annotation results of the plurality of sequenced cells are evaluated; wherein the annotation results are used to indicate the cell type to which each cell belongs, and the cell type to which the cell belongs is the cell type to which the cell population to which the cell belongs belongs.

[0013] Optionally, after evaluating the annotation results of the plurality of sequenced cells, the method further comprises:

[0014] Remove the annotation results that meet the preset filter conditions after evaluation.

[0015] Optionally, the evaluating the annotation results of the plurality of sequenced cells comprises:

[0016] training a cell type classifier using the cell sequencing data and annotation results of a portion of the plurality of cells;

[0017] Processing the cell sequencing data of another portion of the cells in the plurality of cells by the cell type classifier to obtain a classification result of the another portion of the cells; wherein the classification result is used to indicate the cell type to which each cell determined by the cell type classifier belongs;

[0018] An evaluation result of the annotation result is determined according to whether the annotation result of the other part of cells is consistent with the classification result.

[0019] Optionally, the multiple cell populations obtained by clustering the cell sequencing data include:

[0020] Preprocessing the cell sequencing data;

[0021] Clustering is performed based on the preprocessed cell sequencing data to obtain multiple cell populations.

[0022] Optionally, determining the probability value of the cell population belonging to each cell type according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population includes:

[0023] Determining the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components in the cell population;

[0024] Obtaining marker component-cell type marker data; wherein the marker component-cell type marker data represents the probability that the cell belongs to the cell type when the expression level of the marker component in the cell is higher than a preset expression level threshold;

[0025] The probability value of the cell population belonging to each of the cell types is calculated based on the score of the marker component in the cell population and the marker component-cell type marker data corresponding to the marker component.

[0026] Optionally, determining the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components in the cell population includes:

[0027] For each marker component in the cell type-marker component correspondence, the sum of the reciprocal of the ranking of the marker component in the differentially highly expressed component list to N is divided by N to obtain the score of the marker component in the cell population; wherein N is a positive integer;

[0028] or,

[0029] For each marker component in the cell type-marker component correspondence, the reciprocal of the ranking of the marker component in the differentially highly expressed component list is divided by the sum of the reciprocals from 1 to N to obtain the score of the marker component in the cell population.

[0030] A second aspect of the present application provides a device for annotating cell types, comprising:

[0031] A clustering unit, configured to obtain cell sequencing data of a plurality of cells and cluster the obtained cell populations according to the cell sequencing data;

[0032] an analysis unit, configured to obtain a list of differentially highly expressed components of each cell population based on the cell sequencing data; wherein the list of differentially highly expressed components comprises a plurality of differentially highly expressed components in the cell population sorted from high to low according to specificity scores, and the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins;

[0033] An annotation unit is used to determine, for each of the cell populations, a probability value that the cell population belongs to each of the cell types according to the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and determine the cell type of the cell population according to the probability value; wherein the marker component is a marker gene or a marker protein.

[0034] A third aspect of the present application provides an electronic device, including a memory and a processor;

[0035] Wherein, the memory is used to store computer programs;

[0036] The processor is used to execute the computer program, and when the computer program is executed, it is specifically used to implement the cell type annotation method provided in any one of the first aspects of the present application.

[0037] The fourth aspect of the present application provides a computer storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the cell type annotation method provided in any one of the first aspects of the present application.

[0038] The present application provides a method, apparatus, device and storage medium for cell type annotation, the method comprising: obtaining cell sequencing data of multiple cells, and clustering multiple cell populations obtained based on the cell sequencing data; obtaining a list of differentially expressed components of each cell population based on the cell sequencing data analysis; wherein the list of differentially expressed components includes multiple differentially expressed components in the cell population sorted from high to low by specificity score, and the differentially expressed components are differentially expressed genes or differentially expressed proteins; for each cell population, determining the probability that the cell population belongs to each cell type based on the ranking of the marker components corresponding to each cell type in the list of differentially expressed components of the cell population in a preset cell type-marker component correspondence, and determining the cell type with the highest corresponding probability as the cell type to which the cell population belongs; wherein the marker components are marker genes or marker proteins, and the higher the ranking of the marker components corresponding to the cell type in the list of differentially expressed components of the cell population, the greater the probability that the cell population belongs to the cell type. This scheme can enhance the determining effect of marker genes with strong specificity on the cell type to which the cell population belongs, and weaken the determining effect of marker genes with weak specificity on the cell type to which the cell population belongs, thereby improving the accuracy of cell type annotation results. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] FIG1 is a flow chart of a method for cell type annotation provided in an embodiment of the present application;

[0040] FIG2 is a schematic diagram of differentially expressed genes sorted by specificity in a cell population provided in an embodiment of the present application;

[0041] FIG3 is a schematic diagram of a cell type annotation result provided in an embodiment of the present application;

[0042] FIG4 is a schematic diagram of a visualized confusion matrix provided in an embodiment of the present application;

[0043] FIG5 is a schematic diagram of an evaluation result of a cell type annotation result provided in an embodiment of the present application;

[0044] FIG6 is a schematic diagram of another cell type annotation result provided in an embodiment of the present application;

[0045] FIG7 is a schematic diagram of another visualized confusion matrix provided in an embodiment of the present application;

[0046] FIG8 is a schematic diagram of evaluation results of another cell type annotation result provided in an embodiment of the present application;

[0047] FIG9 is a schematic structural diagram of a cell type annotation device provided in an embodiment of the present application;

[0048] FIG10 is a schematic structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0049] The present application provides an imaging method and an imaging system for improving the comprehensiveness of image display.

[0050] The terms "first," "second," "third," "fourth," and the like (if any) in the specification and claims of this application and in the accompanying drawings are used to distinguish similar objects and are not necessarily used to describe a particular order or sequential sequence. It should be understood that the terms used in this manner are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" and "having," and any variations thereof, are intended to cover non-exclusive inclusions, e.g., a process, method, system, product, or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0051] An embodiment of the present application provides a method for annotating cell types, see FIG1 , and the method may include the following steps.

[0052] S101, reading cell sequencing data.

[0053] Cell sequencing data is obtained by sequencing multiple cells in advance using cell sequencing technology. After each cell is sequenced, the cell sequencing data corresponding to the cell will be obtained.

[0054] Depending on the cell sequencing technology used, the content of the cell sequencing data will vary.

[0055] For example, if sequencing is performed using spatial transcriptomics technology, the sequencing data corresponding to each cell may include the expression amount of each gene in the cell, or the expression level of each gene.

[0056] If sequencing is performed using spatial proteomics technology, the sequencing data corresponding to each cell may include the content of proteins expressed by each gene in the cell, or the expression levels of proteins expressed by different genes.

[0057] The cell type annotation method provided in this embodiment has the same execution process for the above two types of cell sequencing data. For ease of understanding, the following mainly uses the cell sequencing data containing gene expression levels as an example for explanation.

[0058] In this embodiment, the expression levels of genes or proteins can be expressed as fold change.

[0059] S102, preprocessing cell sequencing data.

[0060] Preprocessing of cell sequencing data may include:

[0061] 1. Remove abnormal data. Specifically, data in cell sequencing data with expression levels far above or far below a preset expression level range can be identified as abnormal data and removed. For example, if the expression level of a gene in the cell sequencing data of a cell is far above the preset expression level range, the cell sequencing data of the cell is identified as abnormal data and removed.

[0062] Second, fill in missing data. Specifically, if the expression level of a gene is missing in the cell sequencing data of a cell, the cell sequencing data of the cell can be filled in based on the expression level of the gene in other cells.

[0063] In addition to the two pre-processing methods listed above, other methods can also be used for pre-processing. For details, please refer to relevant literature in the field of cell sequencing, which is not limited in this embodiment.

[0064] S103, cell clustering.

[0065] Cell clustering is performed based on the gene expression signature of each cell, which can be determined based on the expression levels of each gene in the cell's sequencing data. For example, the gene expression signature of a cell can be represented as (A1, A2, A3...An), where A1 represents the expression level of gene 1 in the cell, A2 represents the expression level of gene 2 in the cell, and so on.

[0066] An optional cell clustering method may be to perform dimensionality reduction processing on a gene expression matrix composed of gene expression characteristics of multiple cells, and then cluster the principal components in the matrix after dimensionality reduction using a graph clustering algorithm (Leiden algorithm), thereby dividing the large number of cells sequenced in step S101 into multiple cell groups.

[0067] It is understandable that the algorithm used for clustering in step S103 may also be other clustering algorithms different from the Leiden algorithm, for example, it may be a K-means algorithm.

[0068] In addition, iterative clustering may be performed in step S103, that is, clustering may be repeated multiple times, and low-quality clustering results may be filtered out after each clustering, so as to improve the accuracy of the finally obtained clustering result.

[0069] Among them, if the similarity of the gene expression characteristics of each cell in the cell population is low, the cell population can be defined as a low-quality clustering result.

[0070] The above description uses cell sequencing data containing gene expression levels as an example. It is understood that the above method can also be applied to clustering cell sequencing data containing protein expression levels obtained through spatial proteomics sequencing. The specific implementation method is consistent with the previous description and will not be repeated here.

[0071] The process from step S101 to step S103 is equivalent to:

[0072] Acquire cell sequencing data of multiple cells, and cluster multiple cell populations based on the cell sequencing data.

[0073] Wherein, step S102 is equivalent to:

[0074] Preprocess cell sequencing data.

[0075] Step S103 is equivalent to:

[0076] Clustering is performed based on the preprocessed cell sequencing data to obtain multiple cell populations.

[0077] S104, Cell type annotation.

[0078] In step S104, the cell type to which each cell group belongs can be determined based on the multiple cell groups obtained by the previous clustering and the cell sequencing data preprocessed in step S102. This process is called cell type annotation.

[0079] Step S104 may specifically include the following steps:

[0080] S141, obtaining cell type-marker gene correspondence data.

[0081] The above cell type (celltype)-marker gene (marker) correspondence data can be called a celltype-marker dictionary.

[0082] The above correspondence data is used to illustrate which marker genes correspond to each preset cell type.

[0083] In this embodiment, the above correspondence data can be expressed in a variety of forms. For example, the cell type-marker gene correspondence data can be expressed in a matrix form or in a list form.

[0084] Please see Table 1 for an example of cell type-marker gene correspondence data in matrix form.

[0085] Table 1

[0086] In Table 1, rows represent cell types, columns represent marker genes, and the value in each cell is 0 or 1. A cell value of 0 indicates that the marker gene in the column is not expressed (or the expression level is low) in the cell type in the row, and a cell value of 1 indicates that the marker gene in the column is expressed at a high level in the cell type in the row.

[0087] For example, a cell in the CD4 column of the T cell row with a value of 1 indicates that the CD4 marker gene is highly expressed in T cells. A cell in the CD4 column of the B cell row with a value of 0 indicates that the CD4 marker gene is lowly expressed or not expressed in B cells.

[0088] It is understandable that if the cell sequencing data obtained in the previous step includes protein expression levels, then what is obtained in step S141 is the cell type-marker protein correspondence.

[0089] S142, calculating the differentially highly expressed genes of the cell population and sorting them by specificity to form a list of differentially highly expressed genes of the cell population.

[0090] Step S142 is equivalent to obtaining a list of differentially highly expressed components of each cell population based on the cell sequencing data analysis.

[0091] The differentially highly expressed component list includes multiple differentially highly expressed components in the cell population sorted from high to low according to specificity scores, and the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins.

[0092] The differentially highly expressed genes of a cell population refer to genes that have a higher expression level in the cell population (generally defined as an expression level higher than a preset expression level threshold) and have significant differences in the cell population.

[0093] Here, a gene having a significant difference in a cell population can be understood as the expression level of the gene in the cell population having a significant difference relative to the expression levels in other cell populations.

[0094] For genes with significantly different expression levels, expression specificity can be used to measure the degree of difference in the gene. The higher the expression specificity, the greater the difference between the expression level of the gene in this cell population and the expression level in other cell populations.

[0095] The expression level of a gene in a cell population can be defined as the average expression level of the gene in all cells of the cell population.

[0096] The expression specificity of a gene in a cell population can be expressed by a P-value. The larger the P-value, the higher the expression specificity of the gene in the cell population, and vice versa.

[0097] Whether a gene has a significant difference in a cell population, and the P value when there is a significant difference, can be calculated using any significance test method, for example, the Wilcoxon test method or the t-test method. This embodiment does not limit the specific algorithm.

[0098] In summary, in step S142, the following operations may be performed for each cell population:

[0099] First, based on the cell sequencing data of each cell in the cell population, genes with high expression levels and significant differences in the cell population are determined, that is, the differentially highly expressed genes of the cell population are determined. Then, for each differentially highly expressed gene in the cell population, the specificity score of the gene in the cell population is obtained by comprehensively calculating the expression level and expression specificity of the gene in the cell population. Finally, these differentially highly expressed genes are sorted from high to low according to their specificity to form a list of differentially highly expressed genes in the cell population.

[0100] The specific calculation method of the specificity score can be any existing calculation method according to the actual application scenario, and this embodiment does not limit this. As an example, it can be calculated using the Wilcoxon test method.

[0101] [Corrected 18.08.2023 according to Rule 26] Please refer to Figure 2, which is a schematic diagram of the differentially expressed genes sorted by specificity in the cell population provided in the embodiment of the present application. "3 relative to others" indicates that the schematic diagram shows the specificity ranking of the third type of cell population obtained after clustering, wherein the horizontal axis represents the sequence number of each differentially expressed gene sorted from high to low by specificity, and the vertical axis represents the specificity score of the corresponding gene. It can be seen that the Mouse-CD4 gene in the first position (i.e., 0 on the horizontal axis) has the highest specificity score, the Mouse-CD8a in the second position has the second highest specificity score, and so on.

[0102] It is understandable that FIG2 shows the top five differentially highly expressed genes in the differentially highly expressed gene list for simplicity. In actual applications, the differentially highly expressed gene list of a cell population may include more differentially highly expressed genes, not limited to the five shown in FIG2 .

[0103] When the cell sequencing data includes protein expression levels, step S142 may be to calculate the differentially highly expressed proteins of the cell population and sort them by specificity to form a list of differentially highly expressed proteins of the cell population. The specific execution method is consistent with the above and will not be repeated here.

[0104] S143, determining the score of the marker gene in the cell population according to the order of the marker gene in the list of differentially highly expressed genes in the cell population.

[0105] The order of the marker genes in the list of differentially highly expressed genes in each cell population can be represented by the cell population-marker gene matrix shown in Table 2 below.

[0106] Table 2

[0107] In Table 2, rows represent cell populations, columns represent marker genes, and the value in each cell is a non-negative integer. When the value is 0, it means that the marker gene corresponding to the column where the cell is located does not exist in the first N positions of the differentially highly expressed gene list of the cell population in the row where the cell is located; when the value is a positive integer, the value indicates the position of the marker gene corresponding to the column where the cell is located in the differentially highly expressed gene list of the cell population in the row where the cell is located.

[0108] N is a positive integer preset by the user. For example, the matrix shown in Table 2 is when N is set to 5. In other embodiments, N can also be set to other values, such as 30. This embodiment does not limit the specific value of N.

[0109] For example, the cell value in row 0, CD4 column is 4, indicating that the marker gene CD4 is at the 4th position in the list of differentially highly expressed genes of type 0 cell population; the cell value in row 0, CD8 column is 0, indicating that the marker gene CD8 is not in the first N positions of the list of differentially highly expressed genes of type 0 cell population.

[0110] For each cell population, according to the ranking of each marker gene in the differentially highly expressed gene list of the cell population, the effect score of each marker gene in the top N positions of the differentially highly expressed gene list can be calculated using the following formula (1) or formula (2).

[0111] Formula (1):

[0112] Formula (2):

[0113] S i (ROC) represents the effect score of the i-th marker gene in the differentially expressed gene list (hereinafter referred to as the list) calculated by formula (1); S i (RR) represents the effect score of the i-th marker gene calculated using formula (2). i Indicates the order of the i-th marker gene appearing in the list.

[0114] Taking Table 2 as an example, in the list of differentially highly expressed genes of type 0 cell population, the first marker gene that appears is CD4, and the ranking r1 of this marker gene in the list is 4. The second marker gene that appears is IgD, and the ranking r2 of this marker gene in the list is 1.

[0115] The specific calculation process of formula (1) and formula (2) is described below with reference to the example in Table 2.

[0116] Formula (1) indicates that for the i-th marker gene with non-zero ranking in the differentially expressed gene list of each cell population, find the ranking r of the marker gene in the differentially expressed gene list. i , calculate one by one from r i to the reciprocal of N, and then divide the sum of these reciprocals by N to obtain the effect score of the i-th marker gene in the list.

[0117] Combined with the example of the 0th cell population in Table 2, for the first marker gene CD4 of the 0th cell population, its effect score is the reciprocal of the ranking r1=4 plus the reciprocal to N=5, divided by N=5, and the obtained effect score S1(ROC)=0.09. For the second marker gene IgD of the 0th cell population that is non-zero (only the effect scores of genes with non-zero rankings are calculated), it is ranked first in the differentially highly expressed list, that is, r2=1, and its effect score is the reciprocal of the ranking r2=1 plus the reciprocal to N=5, that is, (1 / 1+1 / 2+1 / 3+1 / 4+1 / 5), divided by N=5, and the obtained effect score is S2(ROC)=0.46.

[0118] In other words, the calculation method for the effect score of the marker component on the cell population can be:

[0119] For each marker component in the cell type-marker component correspondence, the score of the marker component in the cell population is obtained by dividing the sum of the reciprocal of N in the list of differentially highly expressed components by N; where N is a positive integer.

[0120] Formula (2) indicates that for the i-th marker gene with a non-zero ranking in the differentially highly expressed list of various cell populations, the reciprocal of the ranking of the marker gene in the list is divided by the sum of the reciprocals from 1 to N to obtain the effect score of the i-th marker gene.

[0121] Still combining with the example of the cell population of type 0 in Table 2, still combining with the example of the cell population of type 0 in Table 2, for the first marker gene CD4 in the list of cell population of type 0, its effect score is the reciprocal of the ranking r1=4, divided by the sum of the reciprocals from 1 to N=5 (1+1 / 2+1 / 3+1 / 4+1 / 5), and its effect score is 0.11; for the second marker gene IgD in the list of differentially highly expressed genes in cell population of type 0, its effect score is the reciprocal of the ranking r2=1, divided by the sum of the reciprocals from 1 to N=5, and its effect score is 0.44.

[0122] In other words, the calculation method for the effect score of the marker component on the cell population can also be:

[0123] For each marker component in the cell type-marker component correspondence, the reciprocal of the marker component's ranking in the differentially highly expressed component list is divided by the sum of the reciprocals from 1 to N to obtain the score of the marker component in the cell population.

[0124] S144, determining the type of the cell population according to the scores of each marker gene in the cell population.

[0125] The specific execution method of step S144 may be:

[0126] First, the effect scores of each marker gene in each cell population are combined into a scoring matrix S. The specific form of the scoring matrix can be shown in Table 3.

[0127] Table 3

[0128] In Table 3, behavioral cell groups are listed as marker genes. The value in each cell represents the marker gene of the column where the cell is located. The ROC formula is used in the cell group in the row, that is, the effect score calculated by the above formula (1).

[0129] For example, the cell value in row 0, CD4 column is 0.09, indicating that the effect score of the CD4 gene in the 0th cell population is 0.09; the cell value in row 0, CD8 column is 0, indicating that there is no marker gene CD8 in the first N positions of the differentially highly expressed gene list of the 0th cell population.

[0130] Then determine the marker matrix L, the behavioral cell types of the marker matrix, and list them as marker genes. The value of each cell in the marker matrix can be set according to user needs.

[0131] As an example, the value of each cell in the marker matrix can be set to 0 or 1. In this case, the marker matrix L can be considered to be equivalent to the cell type-marker gene correspondence data in matrix form obtained in step S141, and the meaning of each cell is also consistent with the aforementioned cell type-marker gene correspondence data in matrix form, and will not be repeated.

[0132] As another example, the value of each cell in the marker matrix can be set to a real number greater than or equal to 0 and less than or equal to 1, that is, it is set to a continuous value between 0 and 1. In this case, the value in each cell can represent the magnitude of the determining effect of the marker gene in its column on the cell type in its row. The larger the value in the cell, the greater the determining effect of the marker gene in the column where the cell is located on the cell type in its row, and the smaller the value in the cell, the smaller the determining effect of the marker gene in the column where the cell is located on the cell type in its row.

[0133] The decisive effect of a specific marker gene on a specific cell type can be understood as follows: when a specific marker gene has a higher expression level in a cell, the possibility that the cell belongs to a specific cell type is greater when the decisive effect is greater, and conversely, the smaller the decisive effect is, the smaller the possibility is.

[0134] Finally, the scoring matrix and the labeling matrix are combined to calculate the annotation matrix P of the cell population using the following formula (3): P = LS T

[0135] That is, the annotation matrix P is equal to the transpose of the label matrix and the score matrix (i.e. S T ).

[0136] In the annotation matrix, rows are cell populations and columns are cell types, where the value in each cell is a real number between 0 and 1. The value in a cell represents the probability that the cell population in the row where the cell is located belongs to the cell type in the column where the cell is located.

[0137] Obviously, after obtaining the annotation matrix, for each cell population, the probability value of the cell population belonging to each cell type can be compared according to the annotation matrix, and then the cell type corresponding to the largest probability can be marked as the type to which the cell population belongs.

[0138] As an example, the obtained annotation matrix can be shown in Table 4:

[0139] Table 4

[0140] According to the example in Table 4, the probabilities that the cell group 0 belongs to T cells, B cells, and monocytes are 0.5, 0.2, and 0.3, respectively. The probability of belonging to T cells is the highest. Therefore, the cell type to which the cell group 0 belongs can be labeled as T cells.

[0141] It can be understood that the specific execution process described in steps S143 and S144 using genes as an example can be directly applied to cell sequencing data containing protein expression levels obtained by spatial proteomics technology. It is only necessary to replace the genes involved in the above process with corresponding proteins, and no further details will be given.

[0142] The specific implementation process of S144 above is equivalent to:

[0143] For each cell population, the score of the marker component in the cell population is determined according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population;

[0144] Obtaining marker component-cell type marker data; wherein the marker component-cell type marker data represents the probability that the cell belongs to a cell type when the expression level of the marker component in the cell is higher than a preset expression level threshold;

[0145] For each cell population, the probability value of the cell population belonging to each cell type is calculated based on the score of the marker component in the cell population and the marker component-cell type marker data corresponding to the marker component.

[0146] Marker component - cell type marker data, equivalent to the value in each cell in the aforementioned marker matrix.

[0147] Step S141, as well as steps S142 and S143, are equivalent to:

[0148] For each cell population, the probability value of the cell population belonging to each cell type is determined based on the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and the cell type to which the cell population belongs is determined based on the probability value of the cell population.

[0149] The method of determining the cell type according to the probability value may be to determine the cell type with the highest corresponding probability as the cell type to which the cell group belongs.

[0150] In addition, the method of determining the cell type based on the probability value can also be to first identify the abnormal probability value among multiple probability values ​​and exclude the abnormal probability value, and then determine the cell type with the largest corresponding probability value among the remaining probability values ​​as the cell type to which the cell population belongs.

[0151] Abnormal probability values ​​can be identified by identifying probability values ​​greater than 1 or less than 0 as abnormal probability values; or by adding the probability values ​​of a cell population belonging to each cell type to obtain a total probability value. If the total probability value is greater than 1 and the total probability value is exactly equal to 1 after removing one of the probability values, the probability value is identified as an abnormal probability value.

[0152] The above method of identifying anomaly probability values ​​is only an example. In other optional embodiments, anomaly probability values ​​may also be identified based on other identification methods.

[0153] Another method for determining the cell type based on the probability value is to, for a cell population, remove the maximum probability value from the probability values ​​of the cell population belonging to a given cell type, and determine the cell type corresponding to the second largest probability value among the original probability values, that is, the cell type corresponding to the largest probability value after exclusion, as the cell type to which the cell population belongs.

[0154] By excluding abnormal probability values ​​or excluding the maximum probability value as described above, the influence of interference factors in the sequencing process on the annotation results can be avoided, thereby improving the accuracy of the annotation results.

[0155] Among them, the marker component is a marker gene or a marker protein. The higher the marker component corresponding to the cell type is ranked in the list of differentially highly expressed components of the cell population, the greater the probability that the cell population belongs to the cell type.

[0156] S105, annotation result evaluation.

[0157] Step S105 is equivalent to:

[0158] The annotation results of the plurality of sequenced cells are evaluated; wherein the annotation results are used to indicate the cell type to which each cell belongs, where the cell type to which the cell belongs is the cell type to which the cell population belongs.

[0159] The method for evaluating the annotation results can be determined according to user needs, and this embodiment does not limit the specific method.

[0160] After step S104, it is possible to determine which cell type each cell group belongs to, and after clustering in step S103, it is determined which cell group each sequenced cell belongs to. Therefore, by combining steps S103 and S104, it is possible to determine which cell type each sequenced cell belongs to, that is, for each cell, the cell type to which the cell belongs is the cell type to which the cell group to which the cell belongs belongs.

[0161] For example, a cell X is clustered into cell group 1, and cell group 1 is annotated as belonging to a macrophage. Therefore, it can be determined that cell X belongs to a macrophage. In other words, the annotation result for cell X is that cell X belongs to a macrophage.

[0162] Step S105 is to evaluate the annotation results of the multiple sequenced cells to determine whether the annotation result of each cell is accurate.

[0163] As an example, a support vector machine (SVM) linear classifier can be used to evaluate the annotation results. Specifically, 80% of the cells of each type can be randomly selected, and the cell sequencing data of these selected cells can be used as a training set to train the SVM classifier. After the training is completed, the cell sequencing data of the remaining 20% ​​of the cells of each type are input into the trained SVM classifier to obtain the classification results of the remaining 20% ​​of the cells. The classification results are then compared with the annotation results of the cell type annotation in step S104, and the accuracy of the annotation results is calculated based on the deviation between the two. The smaller the deviation between the two, the more accurate the annotation results, and the larger the deviation between the two, the less accurate the annotation results.

[0164] Further optionally, when the annotation results are evaluated with an SVM linear classifier, the deviation between the classification results and the annotation results can be represented in the form of a confusion matrix, and the confusion matrix can be visualized to show the quality of the annotation results.

[0165] A confusion matrix, also known as an error matrix, is a standard format for expressing accuracy assessments. In this example, the columns of the confusion matrix represent the number of cells belonging to each cell type in the remaining 20% ​​of cells in the classification results output by the SVM linear classifier, and the rows represent the number of cells belonging to each cell type in the remaining 20% ​​of cells annotated in the annotation results.

[0166] Optionally, if the classification result and annotation result of a cell are determined to be different through the above method, the annotation result of the cell can be removed. For example, after annotation, the cell population to which a cell belongs is determined to be a T cell, but the classification result of the cell sequencing data after classification by the SVM classifier is that the cell belongs to a B cell. In other words, the annotation result (T cell) and the classification result (B cell) of the cell are different. In this case, it can be considered that the accuracy of the cell annotation result is low and may be incorrect. Therefore, the annotation result is removed, that is, the cell is not classified as a T cell.

[0167] The above optional steps are equivalent to removing the annotation results that meet the preset filtering conditions after evaluation.

[0168] In this embodiment, the filtering condition may be that the annotation result and the classification result are different.

[0169] As another example, in step S105, the annotation results may be evaluated by scoring the annotation results based on the class center distance. The specific implementation may be:

[0170] The nearest centroid classifier and multiple five-fold cross validation are used to score the annotation results.

[0171] The number of five-fold cross-validation cycles can be set according to actual conditions. For example, the number of cycles can be set to 1000, that is, 1000 five-fold cross-validation cycles are performed. Finally, the cells are divided into central cells, edge-like cells, and intermediate cells according to the number of accurate predictions for each cell in the final result.

[0172] For example, in the final results, cells with more than 750 accurate predictions are high-quality core cells, cells with less than 250 accurate predictions are low-quality edge cells, and other cells between 250 and 750 are defined as intermediate cells.

[0173] Five-fold cross validation means dividing the cell type annotated dataset, that is, the labeled dataset, into five subsets of equal size, four of which use the nearest centroid classifier to train the model, and the remaining subset is used to verify the model. This process is repeated five times, and a different verification subset is selected each time to ensure that all data is used for training and verification.

[0174] Accurate prediction means that the cell classification result is consistent with the cell annotation result. For example, if a cell is determined to be a T cell through cell type annotation, and the classification result of the cell is also a T cell after processing by the nearest centroid classifier, then the prediction of the cell is considered accurate.

[0175] The process of evaluating the annotation results based on various classifiers in the above example is equivalent to:

[0176] A cell type classifier is trained using cell sequencing data and annotation results of a portion of multiple cells;

[0177] Processing cell sequencing data of another portion of cells in the plurality of cells using a cell type classifier to obtain classification results of the other portion of cells; wherein the classification results are used to indicate the cell type to which each cell determined by the cell type classifier belongs;

[0178] The evaluation results of the annotation results are determined based on whether the annotation results of another part of cells are consistent with the classification results.

[0179] The cell type classifier can be the SVM linear classifier or the nearest centroid classifier mentioned above.

[0180] Optionally, based on the above division, it can be determined that the accuracy of the low-quality edge-cell-like annotation results is low, and the edge-cell-like annotation results are removed.

[0181] The above optional steps are equivalent to removing the annotation results that meet the preset filtering conditions after evaluation.

[0182] In this embodiment, the filtering condition may be low-quality edge-cell-like annotation results.

[0183] That is, the annotation results of low-quality cells that meet the filtering conditions are removed, and the annotation results of non-low-quality cells (i.e., intermediate cells and high-quality cells) are retained.

[0184] The beneficial effects of this embodiment are:

[0185] First, referring to step S102, this embodiment provides different preprocessing methods for the cell sequencing data obtained by spatial proteomics sequencing technology and the cell sequencing data obtained by spatial transcriptomics sequencing technology based on the data characteristics of the spatial proteome and the spatial transcriptome, thereby improving the accuracy of the subsequent annotation results obtained based on the preprocessed cell sequencing data.

[0186] Specifically, spatial transcriptome data typically has thousands to tens of thousands of dimensions and is highly sparse. Therefore, preprocessing of spatial transcriptome data involves 0-1 normalization and logarithmization (log1p) to reduce the impact of high-dimensional sparsity on subsequent analysis. In comparison, spatial proteome data has smaller dimensions, currently only a few hundred at most. Therefore, preprocessing of proteome data involves normalization using the Centered Log-Ratio Normalization (CLR) method, which is more conducive to the discovery of cell populations and type annotation.

[0187] Secondly, according to formula (3), it can be seen that the probability that a cell population in the annotation matrix belongs to a cell type is related to the effect score of the marker gene corresponding to the cell type in the cell population. The greater the effect score of the marker gene corresponding to the cell type in the cell population, the higher the probability that the cell population belongs to the cell type, and vice versa. Further referring to formulas (1) and (2), the effect score of the marker gene in the cell population is related to the ranking of the marker gene in the list of differentially expressed genes in the cell population. The higher the ranking, the stronger the specificity of the marker gene and the higher the corresponding effect score. The lower the ranking, the weaker the specificity of the marker gene and the lower the corresponding effect score.

[0188] Therefore, this scheme can enhance the determining effect of marker genes with strong specificity on the cell type to which the cell population belongs, and weaken the determining effect of marker genes with weak specificity on the cell type to which the cell population belongs, thereby improving the accuracy of cell type annotation results.

[0189] Thirdly, referring to step S105, this algorithm can score the annotation results through an SVM linear classifier or based on the class center distance to achieve quantitative evaluation of the annotation results, which helps to improve the method of cell type annotation through quantitative evaluation.

[0190] Fourthly, this algorithm can remove the annotation results of individual cells with low accuracy based on the evaluation results, and only retain the annotation results with higher accuracy, thereby improving the accuracy of the overall annotation results.

[0191] After using spatial transcriptome technology to sequence a large number of mouse brain cells to obtain cell sequencing data of mouse brain cells, cell type annotation was performed based on the cell type annotation method provided in the embodiments of the present application combined with the cell sequencing data of these mouse brain cells to obtain a spatial distribution diagram of the annotation results as shown in Figure 3.

[0192] Figure 3 shows the cell types of each cell in the mouse brain and the corresponding marker genes for these cell types. The cell types include Hypothalamus neuron, erythrocyte, oligodendrocyte, HPF neuron, thalamus neuron, meninges cell, astrocyte, striatum neuron, CA pyramidal neuron, ependymal cell, choroid plexus epithelial cell, and microglia.

[0193] The annotation results shown in Figure 3 are evaluated using the aforementioned SVM linear classifier to obtain the visualization matrix shown in Figure 4.

[0194] Each row and column in the matrix shown in Figure 4 represents a cell type in Figure 3 and a marker gene corresponding to the cell type. The values ​​in the cells on the diagonal from the upper left corner to the lower right corner of the confusion matrix represent the proportion of cells whose annotation results are consistent with the classification results.

[0195] After evaluating the annotation results shown in Figure 3 using the aforementioned cyclic cross-validation method, a distribution diagram of cells with low-quality annotation results is obtained as shown in Figure 5. It can be seen from Figure 5 that cells with low-quality annotation results are at the edge of each cell group and can be removed to obtain an overall higher-quality annotation result.

[0196] As mentioned above, this embodiment can also perform cell type annotation on cell sequencing data obtained by using spatial proteomics technology.

[0197] After obtaining cell sequencing data from spleen sample cells using spatial proteomics technology, the cell type annotation based on these cell sequencing data was performed using the method provided in this embodiment. The spatial distribution diagram of the obtained annotation results is shown in FIG6 .

[0198] The annotated cell types in Figure 6 include Macrophage, T cell, and B cell.

[0199] The annotation results shown in FIG6 are evaluated using the aforementioned SVM linear classifier to obtain the visualization matrix shown in FIG7 .

[0200] Each row and column in the matrix shown in Figure 7 represents a cell type in Figure 6 and a marker gene corresponding to the cell type. The values ​​in the cells on the diagonal from the upper left corner to the lower right corner of the confusion matrix represent the proportion of cells whose annotation results are consistent with the classification results.

[0201] After evaluating the annotation results shown in Figure 6 using the aforementioned cyclic cross-validation method, a distribution diagram of cells with low-quality annotation results is obtained as shown in Figure 8. It can be seen from Figure 8 that cells with low-quality annotation results are at the edge of each cell group and can be removed to obtain an overall higher-quality annotation result.

[0202] According to the method for cell type annotation provided in this embodiment, this embodiment of the application also provides a device for cell type annotation. Please refer to Figure 9, which is a structural schematic diagram of the device. The device may include the following units.

[0203] A clustering unit 901 is configured to obtain cell sequencing data of a plurality of cells and cluster the obtained cell populations based on the cell sequencing data;

[0204] An analysis unit 902 is configured to obtain a list of differentially highly expressed components for each cell population based on the cell sequencing data analysis; wherein the list of differentially highly expressed components comprises a plurality of differentially highly expressed components in the cell population sorted from high to low by specificity score, wherein the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins;

[0205] Annotation unit 903 is used to determine, for each cell population, the probability value of the cell population belonging to each cell type based on the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and determine the cell type of the cell population based on the probability value; wherein the marker component is a marker gene or a marker protein.

[0206] Optionally, the device further includes:

[0207] The evaluation unit 904 is used to evaluate the annotation results of the plurality of sequenced cells; wherein the annotation results are used to indicate the cell type to which each cell belongs, and the cell type to which the cell belongs is the cell type to which the cell group to which the cell belongs belongs.

[0208] Optionally, after evaluating the annotation results of the plurality of sequenced cells, the evaluation unit 904 is further configured to:

[0209] Remove the annotation results that meet the preset filter conditions after evaluation.

[0210] Optionally, when evaluating the annotation results of the plurality of sequenced cells, the evaluation unit 904 is specifically configured to:

[0211] A cell type classifier is trained using cell sequencing data and annotation results of a portion of multiple cells;

[0212] Processing cell sequencing data of another portion of cells in the plurality of cells using a cell type classifier to obtain classification results of the other portion of cells; wherein the classification results are used to indicate the cell type to which each cell determined by the cell type classifier belongs;

[0213] The evaluation results of the annotation results are determined based on whether the annotation results of another part of cells are consistent with the classification results.

[0214] Optionally, when clustering multiple cell populations obtained by clustering cell sequencing data, the clustering unit 901 is specifically configured to:

[0215] Preprocess cell sequencing data;

[0216] Clustering is performed based on the preprocessed cell sequencing data to obtain multiple cell populations.

[0217] Optionally, the annotation unit 903 determines the probability value of the cell population belonging to each cell type according to the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, specifically for:

[0218] Determining the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the list of differentially highly expressed components in the cell population in the preset cell type-marker component correspondence relationship;

[0219] Obtaining marker component-cell type marker data; wherein the marker component-cell type marker data represents the probability that the cell belongs to a cell type when the expression level of the marker component in the cell is higher than a preset expression level threshold;

[0220] Based on the scores of the marker components in the cell population and the marker component-cell type marker data corresponding to the marker components, the probability values ​​of the cell population belonging to each cell type are calculated.

[0221] Optionally, the annotation unit 903 determines the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, specifically for:

[0222] For each marker component in the cell type-marker component correspondence, the score of the marker component in the cell population is obtained by dividing the sum of the reciprocal of N in the list of differentially highly expressed components by N; where N is a positive integer;

[0223] or,

[0224] For each marker component in the cell type-marker component correspondence, the reciprocal of the marker component's ranking in the differentially highly expressed component list is divided by the sum of the reciprocals from 1 to N to obtain the score of the marker component in the cell population.

[0225] The specific working principle of the cell type annotation device provided in this embodiment can be found in the relevant steps of the cell type annotation method provided in any embodiment of the present application, and will not be repeated here.

[0226] This embodiment provides a device for annotating cell types, including a clustering unit 901 for acquiring cell sequencing data of multiple cells and clustering multiple cell populations based on the cell sequencing data; an analysis unit 902 for obtaining a list of differentially highly expressed components for each cell population based on the cell sequencing data analysis; wherein the list of differentially highly expressed components includes multiple differentially highly expressed components in the cell population ranked from high to low according to specificity scores, and the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins; an annotation unit 903 for determining, for each cell population, the probability that the cell population belongs to each cell type based on the ranking of the marker components corresponding to each cell type in the list of differentially highly expressed components of the cell population in a preset cell type-marker component correspondence, and determining the cell type with the highest corresponding probability as the cell type to which the cell population belongs; wherein the marker component is a marker gene or a marker protein, and the higher the ranking of the marker component corresponding to the cell type in the list of differentially highly expressed components of the cell population, the greater the probability that the cell population belongs to the cell type.

[0227] This scheme can enhance the determining effect of marker genes with strong specificity on the cell type to which a cell population belongs, and weaken the determining effect of marker genes with weak specificity on the cell type to which a cell population belongs, thereby improving the accuracy of cell type annotation results.

[0228] An embodiment of the present application also provides an electronic device, see FIG10 , which includes a memory 1001 and a processor 1002 .

[0229] The memory 1001 is used to store computer programs.

[0230] The processor 1002 is used to execute a computer program. When the computer program is executed, it is specifically used to implement the cell type annotation method provided in any embodiment of the present application.

[0231] An embodiment of the present application also provides a computer storage medium for storing a computer program. When the computer program is executed, it is specifically used to implement the cell type annotation method provided in any embodiment of the present application.

[0232] The above description is merely a specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of this application. Therefore, the scope of protection of this application should be based on the scope of protection of the claims.

Claims

1. A method for annotating cell types, characterized in that: include: Acquire cell sequencing data of multiple cells, and obtain multiple cell populations based on clustering of the cell sequencing data; Obtaining a differentially highly expressed component list for each of the cell populations according to the cell sequencing data analysis; wherein the differentially highly expressed component list comprises a plurality of differentially highly expressed components in the cell population sorted from high to low according to specificity scores, and the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins; For each of the cell populations, the probability value of the cell population belonging to each of the cell types is determined according to the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and the cell type to which the cell population belongs is determined according to the probability value; wherein the marker component is a marker gene or a marker protein.

2. The method according to claim 1, characterized in that Also includes: The annotation results of the plurality of sequenced cells are evaluated; wherein the annotation results are used to indicate the cell type to which each cell belongs, and the cell type to which the cell belongs is the cell type to which the cell group to which the cell belongs belongs.

3. The method according to claim 2, characterized in that After evaluating the annotation results of the plurality of sequenced cells, the method further comprises: Remove the annotation results that meet the preset filter conditions after evaluation.

4. The method according to claim 2, characterized in that: The annotation results of the plurality of sequenced cells are evaluated, including: Using the cell sequencing data and annotation results of a portion of the multiple cells to train a cell type classifier; Processing the cell sequencing data of another part of the cells among the multiple cells by the cell type classifier to obtain a classification result of the other part of the cells; wherein the classification result is used to indicate the cell type to which each cell determined by the cell type classifier belongs; The evaluation result of the annotation result is determined according to whether the annotation result and the classification result of the other part of cells are consistent.

5. The method according to claim 1, characterized in that: The multiple cell populations obtained by clustering the cell sequencing data include: Preprocessing the cell sequencing data; Clustering is performed based on the preprocessed cell sequencing data to obtain multiple cell populations.

6. The method according to claim 1, characterized in that The step of determining the probability value of the cell population belonging to each of the cell types according to the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence relationship in the list of differentially highly expressed components of the cell population comprises: Determine the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence relationship in the list of differentially highly expressed components in the cell population; Obtaining marker component-cell type marker data; wherein the marker component-cell type marker data represents the probability that the cell belongs to the cell type when the expression level of the marker component in the cell is higher than a preset expression level threshold; The probability value of the cell population belonging to each of the cell types is calculated based on the score of the marker component in the cell population and the marker component-cell type marker data corresponding to the marker component.

7. The method according to claim 6, characterized in that The step of determining the score of the marker component in the cell population according to the ranking of the marker component corresponding to each cell type in the preset cell type-marker component correspondence relationship in the list of differentially highly expressed components in the cell population comprises: For each marker component in the cell type-marker component correspondence, the sum of the reciprocals of the ranking of the marker component in the differentially highly expressed component list to N is divided by N to obtain the score of the marker component in the cell population; wherein N is a positive integer; or, For each marker component in the cell type-marker component correspondence, the reciprocal of the ranking of the marker component in the differentially highly expressed component list is divided by the sum of the reciprocals from 1 to N to obtain the score of the marker component in the cell population.

8. A device for annotating cell types, characterized in that: include: A clustering unit, used to obtain cell sequencing data of multiple cells, and cluster multiple cell populations obtained according to the cell sequencing data; An analysis unit is used to obtain a list of differentially highly expressed components of each of the cell populations according to the cell sequencing data; wherein the list of differentially highly expressed components includes the cell populations classified by specificity. A plurality of differentially highly expressed components whose values ​​are ranked from high to low, wherein the differentially highly expressed components are differentially highly expressed genes or differentially highly expressed proteins; An annotation unit is used to determine, for each of the cell populations, the probability value of the cell population belonging to each of the cell types according to the ranking of the marker components corresponding to each cell type in the preset cell type-marker component correspondence in the list of differentially highly expressed components of the cell population, and determine the cell type of the cell population according to the probability value; wherein the marker component is a marker gene or a marker protein.

9. An electronic device, characterized in that: including memory and processor; Wherein, the memory is used to store computer programs; The processor is used to execute the computer program, and when the computer program is executed, it is specifically used to implement the method for cell type annotation as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that: Used to store a computer program, which, when executed, is specifically used to implement the method for cell type annotation as described in any one of claims 1 to 7.

Images