A method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions
By extracting and amplifying cell-associated gene characteristics, predicting cell heterogeneity genes, and combining RAkEL and KEGG pathway analysis, the problems of insufficient accuracy and cell imbalance in endocrine diseases in the prior art were solved, and the problems of insufficient identification of cell function in endocrine diseases were significantly improved. Accuracy of identification of cell functions in endocrine diseases.
Patent Information
- Application Number
- CN202311177280.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-13
- Publication Date
- 2024-11-15
- Estimated Expiration
- 2043-09-13
AI Technical Summary
The existing cell function recognition methods have insufficient prior knowledge and cell imbalance, which makes it difficult to accurately identify the association between endocrine diseases and cell function. The barriers of single-cell sequencing technology and cell number imbalance affect the recognition effect.
By extracting cell-associated gene features, amplifying cell-associated genes, predicting cell heterogeneity genes, and using RAkEL multi-tag multi-classification model and KEGG pathway functional analysis, the cell function of endocrine disease is identified, and classification accuracy and pathway functional recognition are improved.
It improves the accuracy of cell heterogeneity genes and cell function recognition in endocrine diseases, enhances the ability to identify cell functions, and overcomes the problems of small sample gene amplification and label imbalance.
Smart Images

Figure CN117316295B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an endocrine disease cell recognition method, belonging to the technical field of endocrine disease cell function recognition. Background Art
[0002] It has been found that a variety of endocrine cell subtypes are related to major human diseases, such as β cells are closely related to type 2 diabetes, α cells are closely related to type 1 diabetes, and T cells are closely related to polycystic ovary syndrome. The main method for cell function identification and analysis is to analyze the differences in gene expression in different cells through gene sequencing technology. Traditional population sequencing methods provide the average value of gene expression at the level of the entire cell population, but may mask the heterogeneity between cells and the differences between individual cells. Single-cell sequencing technology can obtain detailed information on gene expression at the level of a single cell, thereby being able to analyze the heterogeneity between cells and reveal the differences between different cell types and subtypes.
[0003] At present, the main principle of cell function identification and analysis based on single-cell sequencing technology is to annotate and analyze cells based on differentially expressed genes and known biomarkers. For example, SCDE is a method that uses a Bayesian mixed model to calculate the similarity of differentially expressed gene expression in different cells to identify cell functions. PUSeqCluster is a cell function identification method that uses a joint mixed T distribution clustering model to perform cluster analysis of gene expression data. However, this method still has certain limitations: 1) Due to the lack of prior knowledge, only annotating with known biomarkers and gene expression profiles cannot accurately and effectively identify the association between disease and cell function; 2) Due to the technical barriers of single-cell sequencing and the imbalance of the number of cells related to endocrine tissues, cell heterogeneity gene identification only by analyzing the characteristics of cell function-related genes is affected by cell imbalance.
[0004] Cell heterogeneity function identification is not only related to gene feature extraction, but also analyzing the imbalance of cell-related genes can improve the accuracy of cell classification, thereby effectively identifying cell functions based on endocrine disease pathway functions. Based on machine learning, a cell heterogeneity gene feature amplification method and a disease-related cell function identification method are proposed. A cell function identification method based on cell heterogeneity genes and pathway functions is proposed. This method uses an improved SMOTE algorithm to amplify cell heterogeneity gene features and labels, and uses the RAKEL multi-label multi-classification model to identify cell heterogeneity genes. The cell function identification effect is improved by analyzing the endocrine disease pathway functions involved. Summary of the invention
[0005] The present invention aims to solve the problem of limitations of existing cell function identification methods, and further proposes an endocrine disease cell identification method based on cell heterogeneity genes and pathway functions.
[0006] The technical solution adopted by the present invention to solve the above-mentioned problem is: the steps of the present invention include:
[0007] Step 1, extracting cell-associated gene features;
[0008] Step 2, amplifying cell-associated genes;
[0009] Step 3: Predict cell heterogeneity genes;
[0010] Step 4: Identify endocrine disease cell functions.
[0011] Furthermore, the step of extracting cell-associated gene features in step 1 includes:
[0012] Step 101: Obtain significantly differentially expressed genes in cells based on single-cell sequencing data, and compare them with disease mutation site data and tissue-specific differential expression data based on the location information of the genes. For each gene, select the significant p-values of the five most significantly associated mutation sites as the disease mutation site information of the cell-associated genes;
[0013] Step 102: P-value and log of the matched SNP 2 The FC value was used as the differential expression information of the gene in the tissue, and the eQTL data were derived from pancreas, adipose, blood, and muscle tissues;
[0014] Step 103, the KEGG pathway information associated with the gene is represented as a 343-dimensional binary vector, where each dimension represents the association between the gene and the pathway;
[0015] Step 104: By analyzing the subcellular localization information of cell-associated genes, the subcellular localization information of the top 10 cell substructures representing the genes is selected, and the generative adversarial network is used to generate cell function-related gene features.
[0016] Furthermore, in step 2, the RAkEL framework based on the problem conversion method is used to build an integrated multi-target classification model; RAkEL transforms the multi-target classification problem into a single-label classification problem by treating the label combination of the sample as a new single label. The specific steps include:
[0017] Step 201: Analyze the distribution of gene label combinations, and select small sample genes based on the imbalance ratio IR of the label combination. The definition of IR is as follows:
[0018]
[0019] In formula (1), L represents the label set, L 1 represents the first label, L represents the number of labels, N represents the number of genes, and Y iRepresents the label set corresponding to the i-th gene;
[0020] The average value of IR of all labels, mean(IR), can indicate the imbalance of the data set. It is generally believed that labels with IR(l)>10 can be regarded as small sample labels, and genes containing such labels are called small sample genes.
[0021] Step 202: synthesize gene samples for small sample genes, select a set of k nearest neighbor nodes for each small sample gene, and use Euclidean distance to measure the distance between feature vectors of small sample genes; in order to generate gene labels, count the number of times each label appears in the small sample gene and its neighbor nodes, and set a threshold to synthesize a new label. The label of the synthesized gene is expressed as:
[0022] Label synthGene ={L 1 ,L 2 ,...,L |L|} (2),
[0023] In formula (2), when the number of occurrences of the i-th label in the neighboring node is greater than the set threshold, the label L of the synthetic gene i It is recorded as 1, otherwise it is recorded as 0;
[0024] Step 203: In order to generate gene features, a neighbor node is randomly selected as a reference neighbor gene for generating synthetic sample features, and the gene feature F is synthesized using the interpolation method. syn , which can be expressed as:
[0025] F syn =F seed +r×(F seed -F ref ) (3),
[0026] In formula (3), r is a random number between (0, 1), and the characteristic of the small sample gene is recorded as F seed , refer to the feature representation of the neighboring gene node F ref ;
[0027] Step 204, by comparing the mean(IR) values and similarities of the gene labels and label combination distributions under different small sample gene amplification multiples, the amplification number of small sample genes is selected, which not only improves the balance of gene distribution but also retains the main information of label distribution.
[0028] Furthermore, the step of predicting cell heterogeneity genes in step 3 includes:
[0029] Step 301: divide the labels corresponding to the gene samples into m groups, each group having a label subset of k labels;
[0030] Step 302: construct m binary base classifiers, each classifier performs binary classification on a set of label subsets, and each classifier can make a prediction for its label once: if the i-th classifier P i The corresponding label subset is L i , where each label l j After all classifiers are trained, the average score is calculated for each label l j Final score like If it is greater than the threshold, then the gene is considered to be in the jth cell type Cell j On the contrary, it is considered that the gene has no obvious functional expression in the cell type;
[0031] Step 303: Test the data of the test set on the trained classifiers respectively. Each sample will get a result on each classifier. Finally, the final label of the sample is obtained by voting. The endocrine disease genes are classified into cells. The parameters of RAkEL are k = 5 and m = 14.
[0032] Furthermore, after obtaining the cell-associated gene set in step 4, it is integrated with the cell genes in the original data set into a new cell-associated gene set, and the new gene set is subjected to KEGG pathway-based enrichment analysis to obtain a pathway collection associated with each cell type; the pathway enrichment analysis results obtained by comparing the new cell gene set with the original cell gene set are expressed as in and They respectively represent the gene set identified by the present invention in the i-th cell type and the pathway set obtained after enrichment analysis of the original gene set, which can identify new cell-associated pathways. Since the function of a cell is ultimately reflected in the biological pathway involved in the cell, the cell function is identified by analyzing the endocrine disease pathway function involved in cell heterogeneity genes. The pathway analysis method is implemented using the enrichR package in the R language.
[0033] The beneficial effects of the present invention are as follows: the method first extracts multiple gene features that affect cell functions, and uses a generative adversarial network to generate features of genes related to cell endocrine functions; then, a SMOTE-based gene amplification method is proposed, which improves the problem of low classification accuracy caused by too many small sample genes in the gene label combination by amplifying small sample genes in the gene label combination; finally, a RAkEL-based cell function identification method is proposed, which identifies cell functions by analyzing endocrine disease pathway functions involved in cell heterogeneity genes, and compared with existing similar methods, it is proved that this method effectively improves the accuracy of identifying cell heterogeneity genes and cell functions in endocrine diseases. BRIEF DESCRIPTION OF THE DRAWINGS
[0034] Figure 1 is a flow chart of the present invention;
[0035] Figure 2 It is a schematic diagram comparing the recognition result of the present invention with the benchmark method. DETAILED DESCRIPTION
[0036] Specific implementation method 1: Combination Figure 1 and Figure 2 To illustrate this embodiment, the steps of a method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions described in this embodiment include:
[0037] Step 1, extracting cell-associated gene features;
[0038] Step 2, amplifying cell-associated genes;
[0039] Step 3: Predict cell heterogeneity genes;
[0040] Step 4: Identify endocrine disease cell functions.
[0041] Specific implementation method 2: Combination Figure 1 and Figure 2 This embodiment is described. In step 1 of the method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions, the step of extracting cell-associated gene features comprises:
[0042] Step 101: Obtain significantly differentially expressed genes in cells based on single-cell sequencing data, and compare them with disease mutation site data and tissue-specific differential expression data based on the location information of the genes. For each gene, select the significant p-values of the five most significantly associated mutation sites as the disease mutation site information of the cell-associated genes;
[0043] Step 102: P-value and log of the matched SNP 2 The FC value was used as the differential expression information of the gene in the tissue, and the eQTL data were derived from pancreas, adipose, blood, and muscle tissues;
[0044] Step 103, the KEGG pathway information associated with the gene is represented as a 343-dimensional binary vector, where each dimension represents the association between the gene and the pathway;
[0045] Step 104: By analyzing the subcellular localization information of cell-associated genes, the subcellular localization information of the top 10 cell substructures representing the genes is selected, and the generative adversarial network is used to generate cell function-related gene features.
[0046] Specific implementation method three: Combination Figure 1 and Figure 2 This embodiment is described. In step 2 of the method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions described in this embodiment, an integrated multi-target classification model is constructed using a RAkEL framework based on a problem conversion method. RAkEL converts the multi-target classification problem into a single-label classification problem by treating the label combination of the sample as a new single label. The specific steps include:
[0047] Step 201: Analyze the distribution of gene label combinations, and select small sample genes based on the imbalance ratio IR of the label combination. The definition of IR is as follows:
[0048]
[0049] In formula (1), L represents the label set, L 1 represents the first label, L represents the number of labels, N represents the number of genes, and Y i Represents the label set corresponding to the i-th gene;
[0050] The average value of IR of all labels, mean(IR), can indicate the imbalance of the data set. It is generally believed that labels with IR(l)>10 can be regarded as small sample labels, and genes containing such labels are called small sample genes.
[0051] Step 202: synthesize gene samples for small sample genes, select a set of k nearest neighbor nodes for each small sample gene, and use Euclidean distance to measure the distance between feature vectors of small sample genes; in order to generate gene labels, count the number of times each label appears in the small sample gene and its neighbor nodes, and set a threshold to synthesize a new label. The label of the synthesized gene is expressed as:
[0052] Label synthGene ={L 1 ,L 2 ,...,L |L|} (2),
[0053] In formula (2), when the number of occurrences of the i-th label in the neighboring node is greater than the set threshold, the label L of the synthetic gene i It is recorded as 1, otherwise it is recorded as 0;
[0054] Step 203: In order to generate gene features, a neighbor node is randomly selected as a reference neighbor gene for generating synthetic sample features, and the gene feature F is synthesized using the interpolation method. syn , which can be expressed as:
[0055] F syn =F seed +r×(F seed -F ref ) (3),
[0056] In formula (3), r is a random number between (0, 1), and the characteristic of the small sample gene is recorded as F seed , refer to the feature representation of the neighboring gene node F ref ;
[0057] Step 204, by comparing the mean(IR) values and similarities of the gene labels and label combination distributions under different small sample gene amplification multiples, the amplification number of small sample genes is selected, which not only improves the balance of gene distribution but also retains the main information of label distribution.
[0058] Specific implementation method four: Combination Figure 1 and Figure 2 To illustrate this embodiment, the step of predicting cell heterogeneity genes in step 3 of the endocrine disease cell identification method based on cell heterogeneity genes and pathway functions described in this embodiment includes:
[0059] Step 301: divide the labels corresponding to the gene samples into m groups, each group having a label subset of k labels;
[0060] Step 302: construct m binary base classifiers, each classifier performs binary classification on a set of label subsets, and each classifier can make a prediction for its label once: if the i-th classifier P i The corresponding label subset is L i , where each label l j After all classifiers are trained, the average score is calculated for each label l j Final score like If it is greater than the threshold, then the gene is considered to be in the jth cell type Cell j On the contrary, it is considered that the gene has no obvious functional expression in the cell type;
[0061] Step 303: Test the data of the test set on the trained classifiers respectively. Each sample will get a result on each classifier. Finally, the final label of the sample is obtained by voting. The endocrine disease genes are classified into cells. The parameters of RAkEL are k = 5 and m = 14.
[0062] Specific implementation method five: Combination Figure 1 and Figure 2To illustrate this embodiment, in step 4 of the method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions described in this embodiment, after obtaining the cell-associated gene set, it is integrated with the cell genes in the original data set into a new cell-associated gene set, and the new gene set is subjected to KEGG pathway-based enrichment analysis to obtain a pathway collection associated with each cell type; the pathway enrichment analysis results obtained by comparing the new cell gene set with the original cell gene set are expressed as by analyzing the difference between the two pathway sets. in and They respectively represent the gene set identified by the present invention in the i-th cell type and the pathway set obtained after enrichment analysis of the original gene set, which can identify new cell-associated pathways. Since the function of a cell is ultimately reflected in the biological pathway involved in the cell, the cell function is identified by analyzing the endocrine disease pathway function involved in cell heterogeneity genes. The pathway analysis method is implemented using the enrichR package in the R language.
[0063] The above is only a preferred embodiment of the present invention and does not limit the present invention in any form. Although the present invention has been disclosed as a preferred embodiment as above, it is not used to limit the present invention. Any technician familiar with this profession can make some changes or modify the technical contents disclosed above into equivalent embodiments without departing from the scope of the technical solution of the present invention. However, any simple modification, equivalent replacement and improvement made to the above embodiments without departing from the content of the technical solution of the present invention, based on the technical essence of the present invention, within the spirit and principles of the present invention, still fall within the protection scope of the technical solution of the present invention.
Claims
1. A method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions, characterized by: The method for identifying endocrine disease cells based on cell heterogeneity genes and pathway functions comprises the following steps: Step 1: extracting cell-associated gene features; the step of extracting cell-associated gene features includes: Step 101: Obtain significantly differentially expressed genes in cells based on single-cell sequencing data, and compare them with disease mutation site data and tissue-specific differential expression data based on the location information of the genes. For each gene, select the significant p-values of the five most significantly associated mutation sites as the disease mutation site information of the cell-associated genes; Step 102, using the p-value and log2 FC value of the matched SNP as the differential expression information of the gene in the tissue, and the eQTL data comes from pancreas, fat, blood and muscle tissues; Step 103, the KEGG pathway information associated with the gene is represented as a 343-dimensional binary vector, where each dimension represents the association between the gene and the pathway; Step 104, by analyzing the subcellular localization information of cell-associated genes, selecting the subcellular localization information of the top 10 cell substructures representing genes, and using a generative adversarial network to generate cell function-related gene features; Step 2: Amplify cell-associated genes; use the RAkEL framework based on the problem conversion method to build an integrated multi-target classification model; RAkEL converts the multi-target classification problem into a single-label classification problem by treating the label combination of the sample as a new single label. The specific steps include: Step 201: Analyze the distribution of gene label combinations, and select small sample genes based on the imbalance ratio IR of the label combination. The definition of IR is as follows: In formula (1), L represents the label set, L1 represents the first label, |L| represents the number of labels, N represents the number of genes, and Y i represents the label set corresponding to the i-th gene; the average value of the IR of all labels, mean(IR), can represent the imbalance degree of the data set; it is generally believed that labels with IR(l)>10 can be regarded as small sample labels, and genes containing such labels are called small sample genes; Step 202: synthesize gene samples for small sample genes, select a set of k nearest neighbor nodes for each small sample gene, and use Euclidean distance to measure the distance between feature vectors of small sample genes; in order to generate gene labels, count the number of times each label appears in the small sample gene and its neighbor nodes, and set a threshold to synthesize a new label. The label of the synthesized gene is expressed as: Labelsynth G ene={L1,L2,...,L L }(2), In formula (2), when the number of occurrences of the i-th label in the neighboring node is greater than the set threshold, the label L of the synthetic gene i It is recorded as 1, otherwise it is recorded as 0; Step 203: In order to generate gene features, a neighbor node is randomly selected as a reference neighbor gene for generating synthetic sample features, and the gene feature F is synthesized using the interpolation method. syn , which can be expressed as: F syn =F seed +r×(F seed -F ref )(3), In formula (3), r is a random number between (0, 1), and the characteristic of the small sample gene is recorded as F seed , refer to the feature representation of the neighboring gene node F ref ; Step 204: By comparing the mean(IR) values and similarity of the gene labels and label combination distributions under different amplification multiples of small sample genes, the number of amplifications of small sample genes is selected, which not only improves the balance of gene distribution but also retains the main information of label distribution; Step 3: predicting cell heterogeneity genes; the steps of predicting cell heterogeneity genes include: Step 301: divide the labels corresponding to the gene samples into m groups, each group having a label subset of k labels; Step 302: construct m binary base classifiers, each classifier performs binary classification on a set of label subsets, and each classifier can make a prediction for its label once: if the i-th classifier P i The corresponding label subset is L i , where each label l i After all classifiers are trained, the average score is calculated for each label l i Final score like If it is greater than the threshold, then the gene is considered to be in the jth cell type Cell j On the contrary, it is considered that the gene has no obvious functional expression in the cell type; Step 303: Test the data of the test set on the trained classifiers respectively. Each sample will get a result on each classifier. Finally, the final label of the sample is obtained by voting. The endocrine disease genes are classified into cells. The parameters of RAkEL are k = 5 and m = 14. Step 4: Identify the cell functions of endocrine diseases; after obtaining the cell-associated gene set, integrate it with the cell genes in the original data set into a new cell-associated gene set, and perform KEGG pathway-based enrichment analysis on the new gene set to obtain a collection of pathways related to each cell type; compare the pathway enrichment analysis results obtained from the new cell gene set with the original cell gene set, and analyze the difference between the two pathway sets, expressed as: in and They represent the gene set identified in the i-th cell type and the pathway set obtained after enrichment analysis of the original gene set, which can identify new cell-associated pathways. Since the function of a cell is ultimately reflected in the biological pathways involved in the cell, cell function is identified by analyzing the endocrine disease pathway functions involved in cell heterogeneity genes. The pathway analysis method is implemented using the enrichR package in the R language.
