A single cell identification method based on gene wave and related device

Abstract

This application discloses a single-cell identification method and related devices based on gene waves, relating to the field of single-cell identification technology. The method includes: preprocessing the gene expression matrix of the object to be annotated to obtain a preprocessed matrix; sorting the genes in the preprocessed matrix based on the human reference genome to obtain the gene wave of each single cell to be annotated; using the gene wave of the single cell to be annotated as input, and using a trained single-cell recognition model to determine the identification result of the single cell to be annotated, the identification result being the cell type and / or cell subtype. This application can quickly and accurately complete the automated annotation of cell types and cell subtypes.

Description

Technical Field

[0001] The present application relates to the field of single cell identification technology, and in particular to a single cell identification method based on gene waves and related devices. Background Art

[0002] A major limitation in the current development and evaluation of clinical disease markers, target drugs, or restorative treatments is the lack of rich information about the cellular composition of pathological tissues. Understanding the cellular composition of pathological tissues will fill the knowledge gap that hinders the identification and implementation of effective treatments in this field. Single-cell RNA sequencing (scRNA-seq) technology developed in recent years can provide in-depth understanding of the gene expression patterns and functional states of individual cells. By classifying and annotating individual cells, different cell types can be identified. However, the low-resolution classification level of cell types may not be sufficient to meet the needs. It is necessary to further refine to the high-resolution classification level of cell subtypes to more comprehensively understand the heterogeneity of cell populations. Based on this, different cell subtypes and their gene expression characteristics in healthy and diseased states can be identified, and disease-specific cell subtypes (DSCSs) can be identified, providing strong support for fine-tuning the pathological mechanisms of the disease, discovering potential therapeutic targets, and providing potential therapeutic drugs.

[0003] The accurate definition of cell types and cell subtypes is a prerequisite for the precise identification of DSCSs. Currently, the definition and classification of cell types and cell subtypes are still mainly based on manual annotation. The conventional strategy is to use machine learning methods to perform unsupervised clustering of cell expression profiles. Then, based on the different expression patterns of characteristic genes of different cell types and cell subtypes in each cluster and the functions enriched by highly expressed genes in each cluster, combined with empirical judgment, each cluster identity (i.e., cell type and cell subtype) is manually assigned. Given the lack of clear standards and consensus when assigning identities, different studies have their own subjectivity. At the same time, different studies may use different naming and classification methods for the same cell types and cell subtypes. Manual annotation of DSCSs will lead to increased inconsistency in research results and decreased comparability, thereby limiting the research progress of DSCSs. Therefore, it is extremely urgent to develop tools for automated and accurate annotation of DSCSs.

[0004] To address the challenges posed by manual annotation, multiple studies have developed informatics tools to automatically assign cell identities and have made extensive efforts in automated cell annotation algorithms. These tools are mainly divided into two categories: methods that rely on cell type-specific genes, such as CellAssign, Garnett, and scType; and methods that rely on pre-configured data, such as singleR, scMatch, and scmap. Although these tools provide powerful annotation algorithms at the cell type level, the accuracy of annotating cell types and cell subtypes in complex annotation hierarchies needs to be improved, and the rapid and unified identification of specific cell subtypes cannot be achieved. This has greatly hindered the exploration of key therapeutic targets for diseases and the understanding of related pathogenesis. Therefore, there is an urgent need to develop an automated technology that can quickly and accurately complete the annotation of cell types and cell subtypes. Summary of the Invention

[0005] The purpose of this application is to provide a single-cell identification method and related devices based on gene waves, which can quickly and accurately complete the automated annotation of cell types and cell subtypes.

[0006] To achieve the above objectives, this application provides the following solutions:

[0007] In a first aspect, the present application provides a single-cell identification method based on gene waves, the single-cell identification method based on gene waves comprising:

[0008] Acquire a query data set; the query data set includes a gene expression matrix, the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism;

[0009] The gene expression matrix is ​​preprocessed so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, thereby obtaining a preprocessed matrix; the reference dataset includes a reference gene expression matrix and reference cell metadata, the reference gene expression matrix includes the expression level of each reference gene in each reference single cell included in the annotated object, and the reference cell metadata includes a label of each reference single cell included in the annotated object, wherein the label is a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated;

[0010] Based on the human reference genome, the genes in the preprocessed matrix are sorted, and for each of the single cells to be annotated, the expression levels of the sorted genes in the single cell to be annotated are combined into a gene wave of the single cell to be annotated; the abscissa of the gene wave is the gene, and the ordinate of the gene wave is the expression level of the gene;

[0011] For each of the single cells to be annotated, the gene wave of the single cell to be annotated is used as input, and the identification result of the single cell to be annotated is determined using a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

[0012] In a second aspect, the present application provides a single-cell identification device based on gene waves, the single-cell identification device based on gene waves comprising:

[0013] a data acquisition module for acquiring a query data set; the query data set includes a gene expression matrix, the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism;

[0014] a preprocessing module, configured to preprocess the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, thereby obtaining a preprocessed matrix; the reference dataset comprises a reference gene expression matrix and reference cell metadata, the reference gene expression matrix comprises the expression level of each reference gene in each reference single cell included in the annotated object, the reference cell metadata comprises a label of each reference single cell included in the annotated object, the label being a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated;

[0015] a sorting module, configured to sort the genes in the preprocessed matrix based on the human reference genome, and for each of the single cells to be annotated, compose a gene wave of the single cell to be annotated with the expression levels of the sorted genes of the single cell to be annotated; wherein the abscissa of the gene wave is the gene, and the ordinate of the gene wave is the expression level of the gene;

[0016] The recognition module is used to determine the recognition result of each single cell to be annotated using the gene wave of the single cell to be annotated as input and a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

[0017] In a third aspect, the present application provides a computer device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the above-mentioned single-cell identification method based on gene waves.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned single-cell identification method based on gene waves.

[0019] In a fifth aspect, the present application provides a computer program product, comprising a computer program, which, when executed by a processor, implements the above-mentioned single-cell identification method based on gene waves.

[0020] According to the specific embodiments provided in this application, this application has the following technical effects:

[0021] The present application provides a single-cell identification method and related devices based on gene waves, which preprocess the gene expression matrix of the object to be annotated so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference data set to obtain a preprocessed matrix. Based on the human reference genome, the genes in the preprocessed matrix are sorted to obtain the gene wave of each single cell to be annotated. For each single cell to be annotated, the gene wave of the single cell to be annotated is used as input, and the identification result of the single cell to be annotated is determined using a trained single-cell recognition model. The identification result is a cell type and / or cell subtype. By introducing gene sorting, the present application can consider gene position information in the identification process, and at the same time, the gene wave can be input into the trained single-cell recognition model to complete the identification, thereby quickly and accurately completing the automated annotation of cell types and cell subtypes. Subsequently, based on the cell subtypes obtained by identification in a healthy state and a diseased state, disease-specific cell subtypes are identified. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. 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.

[0023] Figure 1 This is a diagram of the application environment of a single-cell identification method based on gene waves provided in Example 1 of the present application.

[0024] Figure 2 A schematic flow chart of a single-cell identification method based on gene waves provided in Example 1 of the present application.

[0025] Figure 3 Schematic diagram of the principle of a single-cell identification method based on gene waves provided in Example 1 of the present application.

[0026] Figure 4 A schematic diagram of the application process of a single-cell identification method based on gene waves provided in Example 1 of the present application.

[0027] Figure 5 This is a schematic diagram of the gene wave provided in Example 1 of the present application.

[0028] Figure 6 This is a schematic diagram of the network structure of the initial single-cell recognition model provided in Example 1 of the present application; wherein, Figure 6 (a) is the initial single-cell recognition model at the cellular level; Figure 6 (b) is the initial single-cell recognition model at the cell population level.

[0029] Figure 7 A schematic diagram comparing the accuracy of different methods provided in Example 1 of the present application.

[0030] Figure 8 A schematic diagram comparing the mean, median and standard deviation of the accuracy of different methods provided in Example 1 of the present application.

[0031] Figure 9 Schematic diagram of the functional modules of a single-cell identification device based on gene waves provided in Example 2 of the present application.

[0032] Figure 10 A schematic diagram of the structure of a computer device provided in Example 3 of the present application. DETAILED DESCRIPTION

[0033] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0034] Example 1

[0035] The single cell identification method based on gene wave provided in the embodiment of the present application can be applied to Figure 1In the application environment shown. Among them, the terminal communicates with the server through the network. The data storage system can store the data that the server needs to process. The data storage system can be set up separately, integrated on the server, or placed on the cloud or other servers. The terminal can send the query data set to be processed to the server. After the server receives the query data set to be processed, the server preprocesses the gene expression matrix in the query data set for the query data set to be processed, so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference data set, and obtains a preprocessed matrix; based on the human reference genome, the genes in the preprocessed matrix are sorted, and for each single cell to be annotated, the expression amounts of the sorted genes of the single cell to be annotated are combined into the gene wave of the single cell to be annotated; for each single cell to be annotated, the gene wave of the single cell to be annotated is used as input, and the trained single cell recognition model is used to determine the recognition result of the single cell to be annotated. The server can feedback the recognition result of each single cell to be annotated included in the obtained object to be annotated to the terminal.

[0036] In addition, in some embodiments, the single-cell identification method based on gene waves can also be implemented independently by a server or a terminal. For example, the terminal can directly process the query data set to be processed, or the server can obtain the query data set to be processed from the data storage system and process the query data set to be processed.

[0037] The terminals may include, but are not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices may include smart speakers, smart TVs, smart air conditioners, and smart car devices. Portable wearable devices may include smart watches, smart bracelets, and head-mounted devices. The server may be implemented as a standalone server or a server cluster consisting of multiple servers, or as a cloud server.

[0038] In an exemplary embodiment, Figure 2 、 Figure 3 and Figure 4 As shown, a single cell identification method based on gene wave is provided. The method is executed by a computer device, specifically a computer device such as a terminal or a server, or a terminal and a server. In the embodiment of the present application, the method is applied to Figure 1 The following steps are used as an example to illustrate the server.

[0039] Step S1, obtaining a query data set; the query data set includes a gene expression matrix, and the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism.

[0040] Step S2, preprocessing the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, to obtain a preprocessed matrix; the reference dataset includes a reference gene expression matrix and reference cell metadata, the reference gene expression matrix includes the expression level of each reference gene in each reference single cell included in the annotated object, and the reference cell metadata includes the label of each reference single cell included in the annotated object, the label is a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated.

[0041] Step S3, based on the human reference genome, sort the genes in the preprocessed matrix, and for each of the single cells to be annotated, form a gene wave of the single cell to be annotated with the expression levels of the sorted genes; the horizontal axis of the gene wave is the gene, and the vertical axis of the gene wave is the expression level of the gene.

[0042] Step S4: for each of the single cells to be annotated, the gene wave of the single cell to be annotated is used as input, and the identification result of the single cell to be annotated is determined using a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

[0043] By implementing steps S1 to S4 above, this embodiment provides a method for single-cell type and / or subtype identification based on gene waveforms. Inspired by the principles of speech recognition in noisy environments, this method can achieve unified annotation of disease-specific cell types and / or subtypes in heterogeneous single-cell datasets based on the expression waveforms of reordered genes.

[0044] The object to be annotated in this embodiment can be any tissue or organ of an organism of interest to the researcher, such as tendon, muscle, pancreas, lung, liver, etc. After determining the object to be annotated, single-cell transcriptome sequencing is performed on the object to be annotated to obtain the single-cell transcriptome sequencing data of the object to be annotated, that is, to obtain a query dataset. The query dataset includes a gene expression matrix. The gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated. Specifically, it is represented in the form of a gene × cell matrix. Each column of the matrix represents a single cell to be annotated, each row of the matrix represents a gene, and each element of the matrix represents the expression level of a gene in a specific single cell to be annotated.

[0045] Optionally, the query data set of this embodiment may also include cell metadata, which includes additional information of each single cell to be annotated, such as the name of the single cell to be annotated, the tissue to which it belongs, the cell type preliminarily annotated by the user, and the cell subtype preliminarily annotated by the user.

[0046] The above gene expression matrix and cell metadata constitute the query dataset, which is the single-cell transcriptomics data to be annotated provided by the user.

[0047] The present embodiment also designs a reference data set, which is a single-cell transcriptomics data that the user who provides it himself thinks is in line with his needs. The user thinks that the reference data set annotation is correct, and the label of the cell type or cell subtype meets his expectations, which is used as a reference for the gene position on the query data set. Specifically, the reference data set includes a reference gene expression matrix and a reference cell metadata. The reference gene expression matrix includes the expression amount of each reference gene in each reference single cell included in the annotated object, and the reference cell metadata includes the label of each reference single cell included in the annotated object, and the label is cell type and / or cell subtype, that is, the label can be a cell type, and now the annotation of the cell type is completed, the label can be a cell subtype, and now the annotation of the cell subtype is completed, the label can be a cell type and a cell subtype, and now the mixed annotation of the cell type and the cell subtype is completed. The annotated object is identical with the object type to be annotated.

[0048] After introducing the reference dataset, this embodiment further preprocesses the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, obtaining a preprocessed matrix to ensure that the query dataset uses genes that are consistent with the reference dataset. If some reference genes do not exist in the query dataset, they are padded with zeros.

[0049] The gene expression matrix is ​​preprocessed so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, thereby obtaining a preprocessed matrix. Specifically, the preprocessing step includes:

[0050] (1) Compare the genes in the gene expression matrix with the reference genes in the reference gene expression matrix to determine the redundant genes and missing genes. The redundant genes are genes that are present in the gene expression matrix but not in the reference gene expression matrix, and the missing genes are genes that are not present in the gene expression matrix but in the reference gene expression matrix.

[0051] (2) The expression levels of redundant genes in the gene expression matrix are deleted, that is, the rows where the redundant genes are located are deleted, and the missing genes are filled in the gene expression matrix. At the same time, the expression levels of the missing genes are filled with zero, that is, the rows where the missing genes are located are added, and each element in the rows where the missing genes are located is zero, so as to preprocess the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference data set, and the preprocessed matrix is ​​obtained.

[0052] After obtaining the preprocessed matrix, this example further sorts the genes in the preprocessed matrix based on the human reference genome. For each single cell to be annotated, the expression levels of the sorted genes in the single cell to be annotated are combined to form a gene wave for the single cell to be annotated, thereby obtaining a gene wave for each single cell to be annotated. The horizontal axis of the gene wave represents the gene, and the vertical axis represents the gene expression level.

[0053] Among them, the human reference genome specifically uses version number HG38. HG38 (also known as GRCh38, Genome Reference Consortium human genome build 38) is the human genome version released in 2013. HG38 contains the complete sequence information of the human genome, including genes, regulatory regions, repetitive sequences, etc. Based on the human reference genome, the genes in the preprocessed matrix are sorted, specifically including: sorting the genes to be sorted in the preprocessed matrix according to the gene order of the human reference genome HG38, specifically first downloading the hg38.refGene.gtf file of HG38, and after loading, sorting each gene to be sorted in the preprocessed matrix according to the value of the chromosome where it is located and the end site on the chromosome where it is located in ascending order, that is, first sorting according to the value of the chromosome where the gene to be sorted is located in ascending order, and for multiple genes on a chromosome, sorting according to the value of the end site of the gene to be sorted on the chromosome where it is located in ascending order, thereby obtaining the re-sorted genes.

[0054] Among them, after obtaining the reordered genes, only the rows where the reordered genes are located are retained in the preprocessed matrix, and each row in the preprocessed matrix is ​​adjusted according to the reordered genes. Then, each adjusted column is used as the gene wave of the single cell to be annotated corresponding to the column. That is, the gene wave at this time includes the expression level of each reordered gene, and the gene wave of each single cell to be annotated is obtained.

[0055] The genes to be ranked are determined based on the different training methods used by the trained single-cell recognition model. When using a cell population-level training method, the genes to be ranked are the characteristic genes determined during the training process. When using a cell-level training method, the genes to be ranked are all genes in the preprocessed matrix. In this case, when using a cell population-level training method, the genes in the preprocessed matrix are ranked based on the human reference genome, specifically including: ranking all characteristic genes in the preprocessed matrix based on the human reference genome to obtain ranked genes; when using a cell-level training method, the genes in the preprocessed matrix are ranked based on the human reference genome, specifically including: ranking all genes in the preprocessed matrix based on the human reference genome to obtain ranked genes.

[0056] It should be noted that if the preprocessed matrix does not contain the corresponding characteristic genes, the preprocessed matrix is ​​deleted and padded with zeros so that the preprocessed matrix only contains all the characteristic genes.

[0057] This example reorders the genes expressed in each cell according to the position of each gene's termination site in the human reference genome HG38 from smallest to largest, then plots the gene expression levels to obtain their respective gene waves. Gene waves are very similar to the amplitude envelope of sound waves. Each gene can be considered a frame segment, and the gene expression level can be considered the value on the amplitude envelope of the sound wave. Deep learning algorithms for sound wave recognition can then be used to analyze the gene expression of individual cells. Existing cell annotation algorithms mostly directly analyze or learn from the cell-gene expression matrix. This example, based on the principles of deep learning for sound wave recognition, treats the positional information of gene open expression as the temporal information of the sound wave, and each gene interval as a frame segment of the sound wave. ScRNA-seq data is converted into one-dimensional data for learning and annotation according to the order of the human reference genome. This adds a new feature compared to other algorithms: gene location information. It is well known that DNA is arranged in an orderly sequence on each chromosome. The histones in the nucleosome, acting as the spools around which DNA is wound, can influence gene transcription through post-translational modifications at specific sites. Moreover, genes are clustered on nucleosomes. Genes on the same nucleosome may be co-regulated and show similar expression patterns, forming a clustered expression phenomenon. Therefore, after sorting the expressed genes according to the human reference genome, there may be correlation between adjacent genes. Therefore, adding position features is meaningful for each cell. Adding this dimension of position features to the algorithm reshapes the gene expression pattern into a one-dimensional form of gene waves, adds new features, and is closer to the form of gene expression. In principle, it will have a certain effect on improving the accuracy of cell annotation. Figure 5 As shown, the black line is the expression curve of each gene, the red line is the expression curve processed by the Gaussian filter, the gene wave is the black line part, and the red line is a state that is easier to understand.

[0058] After obtaining the gene wave for each single cell to be annotated, the gene wave of each single cell to be annotated is used as input to determine the recognition result of the single cell to be annotated using a trained single-cell recognition model. The trained single-cell recognition model is a model trained based on a reference dataset. When the label is cell type, the recognition result is cell type; when the label is cell subtype, the recognition result is cell subtype; when the label is cell type and cell subtype, the recognition result is cell type and cell subtype.

[0059] The reference dataset is introduced to train the initial single-cell recognition model to obtain a trained single-cell recognition model. The trained single-cell recognition model is then used to complete the recognition of the query dataset. Before using the gene wave of the single cell to be annotated as input and determining the recognition result of the single cell to be annotated using the trained single-cell recognition model, the single-cell recognition method based on gene wave of this embodiment further includes:

[0060] (1) Obtain a reference dataset.

[0061] In the single-cell data, for the reference dataset, genes with a total expression value less than 2 (if less than 2, it may be a misdetection) are deleted, and labels (cell types and / or cell subtypes) with fewer than 10 cells (if less than 10, the corresponding features cannot be extracted) are deleted, because too few cells will cause the trained model to overfit.

[0062] At this point, obtain the reference dataset, including:

[0063] 1) Obtaining the original dataset, which includes the original gene expression matrix and original cell metadata. The original gene expression matrix includes the expression level of each original gene in each original single cell included in the annotated object. The original cell metadata includes the label of each original single cell included in the annotated object and may also include the name of each original single cell included in the annotated object.

[0064] 2) For each original gene and each tag, determine whether the total expression of the original gene is less than a first preset value. If so, record the original gene as a gene to be deleted; determine whether the number of original single cells belonging to the tag is less than a second preset value. If so, record the tag as a tag to be deleted.

[0065] Among them, if the label is cell type, each label refers to each cell type; if the label is cell subtype, each label refers to each cell subtype; if the label is cell type and cell subtype, each label refers to each cell type and each cell subtype.

[0066] The total expression of the original gene is the sum of the expression of the original gene in each original single cell, which is equivalent to the sum of the elements in the row of the original gene in the original gene expression matrix. The first preset value can be 2, and the second preset value can be 10. It should be noted that the original genes other than the gene to be deleted are the reference genes, and the original single cells other than the original single cells with the label to be deleted are the reference single cells.

[0067] 3) Delete the expression amount of the gene to be deleted and the expression amount of each original gene in the original single cell belonging to the label to be deleted in the original gene expression matrix, that is, delete the row where the gene to be deleted and the column where the original single cell belonging to the label to be deleted in the original gene expression matrix to obtain a reference gene expression matrix.

[0068] 4) Based on each reference single cell included in the reference gene expression matrix, the original cell metadata is processed, that is, only the label of each reference single cell is retained in the original cell metadata to obtain the reference cell metadata.

[0069] 5) The reference gene expression matrix and reference cell metadata are combined into a reference dataset.

[0070] (2) Use the reference data set to train the initial single-cell recognition model to obtain a trained single-cell recognition model.

[0071] This embodiment provides two training methods, one is a training method at the cell group level, that is, a training method for single-cell identification based on gene waves from the cell group to the cell level, and the other is a training method at the cell level, that is, a training method for single-cell identification based on gene waves from the cell to the cell level. After the training is completed, the subsequent cell-to-cell level single-cell identification method based on gene waves can be prepared to run.

[0072] If a cell population-level training method is used, the FindAllmarkers() function in the Seurat package is used to calculate the characteristic genes of each cell population (i.e., cell type and / or cell subtype). These characteristic genes are then input into the initial single-cell recognition model at the cell population level for training according to the order of the human reference genome.

[0073] If a cellular-level training method is used, the reference genes are directly input into the initial single-cell recognition model at the cellular level for training according to the order of the human reference genome.

[0074] The initial single cell recognition model can adopt any existing neural network model. As an example, Figure 6 The specific options are as follows:

[0075] If a cellular-level training method is used, based on the gene wave of a single cell, a one-dimensional convolutional neural network (1D CNN) is used. It contains an input layer, five hidden layers, and an output layer connected in sequence. The five hidden layers are 2 convolutional layers, 1 pooling layer, 1 Flatten layer (flat layer), and 1 fully connected layer. That is, the first hidden layer is a convolutional layer, the second hidden layer is a convolutional layer, the third hidden layer is a pooling layer, the fourth hidden layer is a Flatten layer, and the fifth hidden layer is a fully connected layer. The number of reference genes in the reference dataset is used as the dimension of the model input, the number of convolution kernels in the convolution layer is 16, the convolution kernel size is 3, and the Softmax function is used in the fully connected layer.

[0076] If a cell group-level training method is used, a back propagation (BP) neural network is used based on the gene wave of the cell group.

[0077] The training process of the initial single-cell recognition model is mainly divided into two stages. The first stage is the forward propagation of the signal, and the second stage is the back propagation of the error. The steepest descent method is used as the learning rule. The weights and thresholds of the initial single-cell recognition model are continuously adjusted through back propagation to minimize the sum of squared errors of the initial single-cell recognition model.

[0078] If a cell population-level training method is used, the initial single-cell recognition model is trained using a reference dataset to obtain a trained single-cell recognition model, specifically including:

[0079] 1) Feature extraction is performed based on the reference gene expression matrix to determine the characteristic genes corresponding to each label.

[0080] Specifically, the FindAllmarkers() function in the Seurat package can be used to complete the feature extraction process. The reference gene expression matrix and the corresponding reference metadata are used as input. After using the Seurat package to construct the corresponding Object, the FindAllmarkers() function is used to determine the characteristic genes corresponding to each label. Each label corresponds to a cell population, thereby determining the characteristic genes related to the cell population.

[0081] If the label is cell type, the characteristic genes of each cell type can be determined; if the label is cell subtype, the characteristic genes of each cell subtype can be determined; if the label is cell type and cell subtype, the characteristic genes of each cell type and the characteristic genes of each cell subtype can be determined.

[0082] 2) Based on the human reference genome, all characteristic genes are sorted, and for each reference single cell, the expression levels of the sorted characteristic genes of the reference single cell are combined into the gene wave of the reference single cell.

[0083] 3) Using the gene wave of each reference single cell as input and the label of each reference single cell as output, the initial single cell recognition model is trained to obtain a trained single cell recognition model.

[0084] If a cell-level training method is used, the initial single-cell recognition model is trained using a reference dataset to obtain a trained single-cell recognition model, specifically including:

[0085] 1) Based on the human reference genome, the reference genes in the reference gene expression matrix are sorted, and for each reference single cell, the expression levels of the sorted reference genes in the reference single cell are combined into a gene wave of the reference single cell.

[0086] 2) Using the gene wave of each reference single cell as input and the label of each reference single cell as output, the initial single cell recognition model is trained to obtain a trained single cell recognition model.

[0087] When the label is cell type, the trained single-cell recognition model can be used to generate cell type recognition results. When the label is cell subtype, the trained single-cell recognition model can be used to generate cell subtype recognition results. When the label is cell type and cell subtype, two single-cell recognition models are trained, one for generating cell type recognition results and the other for generating cell subtype recognition results.

[0088] Optionally, in this embodiment, the reference dataset may be standardized and normalized before training, and the query dataset may be standardized and normalized before recognition.

[0089] After training is complete, this example uses metrics such as precision, recall, accuracy, F1-score, and receiver operating curves to evaluate the trained single-cell recognition model. Precision refers to the proportion of samples predicted as positive among samples that are truly positive, recall refers to the proportion of samples predicted as positive among all positive samples, and accuracy is a performance metric for evaluating classification problems. For a given set of data, the proportion of correctly classified samples to the total number of samples. F1-score is the harmonic mean of precision and recall and is a measurement metric for classification problems. The receiver operating curve uses the false positive rate as the horizontal axis and the true positive rate as the vertical axis. The area under the curve can be used to comprehensively evaluate the classification performance of the model.

[0090] The training methods at the cell level and the cell group level each have their own characteristics. The cell-to-cell level is more refined, while the cell group-to-cell level can capture group characteristics. Users can use the trained single-cell recognition models obtained by the two training methods for recognition, respectively, and choose the result that is more convincing to them, or make a final choice after comparing the two.

[0091] In this example, all scRNA-seq data (i.e., reference dataset and query dataset) were preprocessed using R (version 3.6.1).

[0092] In this embodiment, the reference data set can be designed as a data set obtained in a healthy state, and the query data set can be designed as a data set obtained in a certain disease state. After completing the cell type / subtype annotation of each single cell in the query data set, based on the proportion of each cell type / subtype in a healthy state (that is, the proportion of the number of single cells of this cell type / subtype to the total number of single cells) and the proportion of each cell type / subtype in a certain disease state, the cell type / subtype whose proportion changes significantly (which may be greater than a preset value) can be regarded as a disease-specific cell type / subtype.

[0093] This embodiment can obtain the cell subtype or type label obtained after annotating each cell, and realize the unified annotation of specific cell types or subtypes for heterogeneous single-cell datasets across diseases. By introducing gene waves and trained single-cell recognition models, the recognition accuracy is higher. This embodiment compares the method used in this embodiment with tools such as Azimuth and Garnett on single-cell data from tissues such as Blood, Brain, Liver, and Pancreas in the HCL (Human Cell Landscape) database. Figure 7 and Figure 8 As shown, Figure 7In the figure, the number after each tissue type represents the number of single cells included in the tissue. For example, "human blood 9649" means that human blood tissue includes 9649 single cells. "our-method-cell" represents the corresponding recognition method when this embodiment adopts the cell-level training method. "our-method-group" represents the corresponding recognition method when this embodiment adopts the cell-level training method. From the average, median, and standard deviation of the accuracy of each data set, it can be seen that the average accuracy (91.15%) and median (97.16%) of the method used in this embodiment are higher than those of other tools (such as the average accuracy (67.95%) and median (89.51%) of Azimuth), and the standard deviation (0.1959) is lower than that of other tools such as Azimuth, indicating that the method of this embodiment has a high accuracy, small fluctuations, and stable performance.

[0094] Current tools can achieve cross-dataset cell type annotation and cell subtype annotation within a dataset. For example, algorithms such as scDeepsort and SingleR can annotate unknown cells to the large subpopulation level based on the information of cells in large databases (HCL, MAC, etc.); algorithms such as scType and CellAssign can complete cell type annotation based on existing cell marker gene databases; algorithms such as scmap and Cellid can complete cell type mapping between small datasets; and the devcellpy algorithm can complete cell subtype annotation on cardiac development models. However, existing tools cannot accurately annotate cell types and cell subtypes across datasets. This may be caused by factors such as batch effects between datasets and missing cell subtype marker genes. Therefore, there are still problems in the rapid and unified annotation of DSCSs across datasets, and more suitable annotation algorithms need to be developed. The use of gene waves in this embodiment solves the problem of annotation failure caused by missing marker genes. The use of neural networks reduces the impact of noise on cell gene expression characteristics and improves the problems caused by batch effects. The development of this embodiment is inspired by the application of deep learning in sound recognition. Just as everyone has a unique voiceprint, each cell has its own unique one-dimensional gene expression pattern: a feature that contains gene expression pattern and position information. As is well known, DNA is ordered and arranged in clusters on each chromosome, and the genes on the same nucleosome may be co-regulated, showing similar expression patterns, forming a phenomenon of clustered expression, therefore, incorporating gene position information is of great significance for the accurate comparison of expression patterns between cells. Simultaneously, with reference to the application of filters in eliminating noise, filtering out unnecessary frequencies, etc., a filtering principle is introduced, and the position information of gene open expression is regarded as the time information of sound waves, and each gene interval is regarded as the frame segment of sound waves, and this point can better predict and supplement missing values ​​in the processing of single-cell data with higher dropout event rates, weaken the impact caused by missing information, thereby more effectively improving the precision of cell annotations. In summary, compared to the traditional annotation method that only considers expression amount, the algorithm based on gene expression position and expression amount information proposed in this embodiment can more effectively improve the precision of cell annotations in principle.

[0095] This example proposes a method for single-cell type and subtype identification based on cell waves, using R and Python software packages. This method enables unified mapping of DSCSs obtained from any tissue, potentially with varying annotation levels. Based on the principle of sound recognition in noisy environments, the method first reorders the genes of each cell according to their position in HG38, converting the expression profile of each gene in the cell into a gene wave format. Then, using a neural network approach, incorporating filter principles, the impact of background noise on DSCS identification in single-cell analysis is reduced, ultimately achieving rapid, unified, and accurate annotation of DSCSs.

[0096] The present application also provides an application scenario, which applies the above-mentioned single-cell identification method based on gene waves. Specifically, the single-cell identification method based on gene waves provided in this embodiment can be applied in a cell annotation scenario. The cell annotation scenario includes a single-cell sequencing link and a single-cell annotation link. The single-cell sequencing link is used to sequence the object to be annotated to obtain single-cell transcriptome sequencing data, and the single-cell annotation link is used to process the single-cell transcriptome sequencing data to obtain the identification results of each single cell to be annotated included in the object to be annotated. The single-cell identification method based on gene waves provided in this embodiment belongs to the single-cell annotation link.

[0097] Example 2

[0098] Based on the same inventive concept, the present application also provides a gene wave-based single-cell identification device for implementing the gene wave-based single-cell identification method mentioned above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more gene wave-based single-cell identification device embodiments provided below can be found in the above-mentioned limitations of the gene wave-based single-cell identification method, and will not be repeated here.

[0099] In an exemplary embodiment, Figure 9 As shown, a single cell identification device based on gene waves is provided, and the single cell identification device based on gene waves includes:

[0100] The data acquisition module M1 is used to obtain a query data set; the query data set includes a gene expression matrix, and the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism.

[0101] The preprocessing module M2 is used to preprocess the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset to obtain a preprocessed matrix; the reference dataset includes a reference gene expression matrix and reference cell metadata, the reference gene expression matrix includes the expression level of each reference gene in each reference single cell included in the annotated object, and the reference cell metadata includes the label of each reference single cell included in the annotated object, the label is a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated.

[0102] The sorting module M3 is used to sort the genes in the preprocessed matrix based on the human reference genome, and for each of the single cells to be annotated, the expression levels of the sorted genes of the single cell to be annotated are combined into a gene wave of the single cell to be annotated; the horizontal axis of the gene wave is the gene, and the vertical axis of the gene wave is the expression level of the gene.

[0103] The recognition module M4 is used to determine the recognition result of each single cell to be annotated using the gene wave of the single cell to be annotated as input and a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

[0104] Example 3

[0105] In an exemplary embodiment, a computer device is provided. The computer device may be a server or a terminal. The internal structure diagram thereof may be as follows: Figure 10 As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O) and a communication interface. The processor, memory and input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a single-cell identification method based on gene waves is implemented.

[0106] Those skilled in the art will understand that Figure 10 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0107] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the single-cell identification method based on gene waves in Example 1 is implemented.

[0108] Example 4

[0109] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, which, when executed by a processor, implements the single-cell identification method based on gene waves in Example 1.

[0110] Example 5

[0111] In an exemplary embodiment, a computer program product is provided, comprising a computer program, which implements the gene wave-based single-cell identification method in Example 1 when executed by a processor.

[0112] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0113] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0114] This document uses specific examples to illustrate the principles and implementation methods of this application. The description of the above examples is only intended to help understand the method and core concept of this application. At the same time, for those skilled in the art, based on the concept of this application, there may be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting this application.

Claims

1. A single cell identification method based on gene waves, characterized in that: The single cell identification method based on gene wave includes: Acquire a query data set; the query data set includes a gene expression matrix, the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism; The gene expression matrix is ​​preprocessed so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, thereby obtaining a preprocessed matrix; the reference dataset includes a reference gene expression matrix and reference cell metadata, the reference gene expression matrix includes the expression level of each reference gene in each reference single cell included in the annotated object, and the reference cell metadata includes a label of each reference single cell included in the annotated object, wherein the label is a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated; Based on the human reference genome, the genes in the preprocessed matrix are sorted, and for each of the single cells to be annotated, the expression levels of the sorted genes in the single cell to be annotated are combined into a gene wave of the single cell to be annotated; the abscissa of the gene wave is the gene, and the ordinate of the gene wave is the expression level of the gene; For each of the single cells to be annotated, the gene wave of the single cell to be annotated is used as input, and the identification result of the single cell to be annotated is determined using a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

2. The single cell identification method based on gene wave according to claim 1, characterized in that: Preprocessing the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference data set to obtain a preprocessed matrix, specifically comprising: Comparing the genes in the gene expression matrix with the reference genes in the reference gene expression matrix to determine redundant genes and missing genes; the redundant genes are genes that are present in the gene expression matrix but not in the reference gene expression matrix, and the missing genes are genes that are not present in the gene expression matrix but present in the reference gene expression matrix; The expression levels of redundant genes in the gene expression matrix are deleted, and the missing genes are filled in the gene expression matrix, and the expression levels of the missing genes are filled with zero, so as to preprocess the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference data set, thereby obtaining a preprocessed matrix.

3. The single cell identification method based on gene wave according to claim 1, characterized in that: Before determining the identification result of the single cell to be annotated using the trained single cell identification model using the gene wave of the single cell to be annotated as input, the single cell identification method based on the gene wave further includes: Acquiring the reference dataset; The reference data set is used to train the initial single-cell recognition model to obtain a trained single-cell recognition model.

4. The single cell identification method based on gene wave according to claim 3, characterized in that: Acquiring the reference dataset specifically includes: Acquire an original data set; the original data set includes an original gene expression matrix and original cell metadata, the original gene expression matrix includes the expression level of each original gene in each original single cell included in the annotated object, and the original cell metadata includes the label of each original single cell included in the annotated object; For each original gene and each tag, determine whether the total expression level of the original gene is less than a first preset value. If so, record the original gene as a gene to be deleted; determine whether the number of original single cells belonging to the tag is less than a second preset value. If so, record the tag as a tag to be deleted; Deleting the expression amount of the gene to be deleted and the expression amount of each original gene in the original single cell belonging to the tag to be deleted from the original gene expression matrix to obtain a reference gene expression matrix; Based on each reference single cell included in the reference gene expression matrix, processing the original cell metadata to obtain reference cell metadata; The reference gene expression matrix and the reference cell metadata are combined into a reference dataset.

5. The single cell identification method based on gene wave according to claim 3, characterized in that: When using the reference dataset to train the initial single-cell recognition model, a cell population-level training method or a cell-level training method is adopted; When a cell population-level training method is used, the initial single-cell recognition model is trained using the reference dataset to obtain a trained single-cell recognition model, specifically including: Perform feature extraction based on the reference gene expression matrix to determine the characteristic genes corresponding to each tag; Based on the human reference genome, all the characteristic genes are sorted, and for each of the reference single cells, the expression levels of the sorted characteristic genes of the reference single cell are combined to form a gene wave of the reference single cell; The initial single-cell recognition model is trained using the gene wave of each reference single cell as input and the label of each reference single cell as output to obtain a trained single-cell recognition model. When a cell-level training method is used, the initial single-cell recognition model is trained using the reference dataset to obtain a trained single-cell recognition model, specifically including: Based on the human reference genome, the reference genes in the reference gene expression matrix are sorted, and for each of the reference single cells, the expression amounts of the sorted reference genes of the reference single cell are combined to form a gene wave of the reference single cell; The gene wave of each reference single cell is used as input, and the label of each reference single cell is used as output to train the initial single cell recognition model to obtain a trained single cell recognition model.

6. The single cell identification method based on gene wave according to claim 5, characterized in that: When a cell population-level training method is used, the genes in the preprocessed matrix are sorted based on the human reference genome, specifically including: Based on the human reference genome, all characteristic genes in the preprocessed matrix are sorted to obtain sorted genes; When a cellular level training method is used, the genes in the pre-processed matrix are sorted based on the human reference genome, specifically including: All genes in the preprocessed matrix are sorted based on the human reference genome to obtain sorted genes.

7. A single cell identification device based on gene waves, characterized in that: The single-cell identification device based on gene waves includes: a data acquisition module for acquiring a query data set; the query data set includes a gene expression matrix, the gene expression matrix includes the expression level of each gene in each single cell to be annotated included in the object to be annotated; the object to be annotated is an organ or tissue of an organism; a preprocessing module, configured to preprocess the gene expression matrix so that the genes in the gene expression matrix are completely consistent with the reference genes in the reference dataset, thereby obtaining a preprocessed matrix; the reference dataset comprises a reference gene expression matrix and reference cell metadata, the reference gene expression matrix comprises the expression level of each reference gene in each reference single cell included in the annotated object, the reference cell metadata comprises a label of each reference single cell included in the annotated object, the label being a cell type and / or cell subtype, and the annotated object is of the same type as the object to be annotated; a sorting module, configured to sort the genes in the preprocessed matrix based on the human reference genome, and for each of the single cells to be annotated, compose a gene wave of the single cell to be annotated with the expression levels of the sorted genes of the single cell to be annotated; wherein the abscissa of the gene wave is the gene, and the ordinate of the gene wave is the expression level of the gene; The recognition module is used to determine the recognition result of each single cell to be annotated using the gene wave of the single cell to be annotated as input and a trained single cell recognition model; the trained single cell recognition model is a model trained based on the reference data set.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the single-cell identification method based on gene waves according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the single-cell identification method based on gene waves described in any one of claims 1 to 6 is implemented.

10. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the single-cell identification method based on gene waves described in any one of claims 1 to 6 is implemented.

Images