Intelligent analysis method and system for heterogeneity of endothelial cells in atherosclerosis

Abstract

The invention discloses a method and a system for intelligently analyzing heterogeneity of endothelial cells in atherosclerosis. The method comprises the following steps: acquiring sample data of the endothelial cells in the atherosclerosis of a patient; performing feature extraction on the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; dividing the endothelial cells into different subgroups according to gene expression characteristics of the endothelial cells and gene expression similarity so as to determine marker genes for distinguishing the different subgroups, and defining and representing characteristics of each subgroup by analyzing expression modes of the marker genes; gene expression differences among different endothelial cell subgroups are compared, differential expression genes of the related endothelial cell subgroups, which are significantly up-regulated or down-regulated in the atherosclerosis process, are identified, and the differential expression genes are used as key molecular markers for understanding functional changes and heterogeneity of endothelial cells. According to the method, the heterogeneity of the endothelial cells in the atherosclerosis can be accurately analyzed.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to an intelligent analysis of endothelial cell heterogeneity in atherosclerosis. Background Art

[0002] In recent years, in the field of atherosclerosis research, traditional analysis methods mainly rely on bulk RNA sequencing (bulk RNA-seq) technology, which comprehensively analyzes the gene expression of whole tissue samples to reveal the overall gene expression changes in the disease state. However, this method is based on the average expression level of multiple cells mixed together, which makes it difficult to capture the differences between individual cells and cannot effectively reflect the heterogeneity characteristics within the cell population, especially the endothelial cell population. In addition, traditional analysis methods usually focus on the detection of known markers or preset target genes, which greatly limits the discovery of new subpopulations and potential biomarkers.

[0003] Atherosclerosis is a highly complex vascular disease whose pathological mechanism involves multiple cell types and their interactions. Among them, endothelial cells, as key regulatory factors, play an important role in the occurrence and development of the disease. However, gene expression analysis based on tissue level is difficult to clearly reveal the functional changes and molecular characteristics of specific endothelial cell subpopulations in the course of the disease, resulting in insufficient understanding of the pathological mechanism. Existing studies have begun to try to use single-cell sequencing technology to analyze cell diversity in complex tissues, providing a new means to reveal the heterogeneity of endothelial cells in atherosclerosis. However, the current technology has significant limitations: the processing and analysis process of single-cell data is complex, and an efficient and systematic method has not yet been formed, making it difficult to accurately identify and characterize endothelial cell subpopulations and their specific roles in the dynamic changes of the disease.

[0004] In particular, for single-cell sequencing studies of atherosclerosis-related endothelial cells, existing technologies are mostly focused on preliminary explorations and lack targeted and high-resolution analysis strategies. The significant heterogeneity within the endothelial cell population, such as the differences between different subpopulations at the functional and molecular levels, is often overlooked. This deficiency directly limits the comprehensive understanding of the role of endothelial cells in the process of atherosclerosis and hinders the discovery of potential molecular markers and new therapeutic targets.

[0005] Therefore, existing technical solutions are mainly limited to gene expression analysis at the tissue level or preliminary single-cell sequencing attempts, and lack the ability to accurately analyze the heterogeneity of atherosclerotic endothelial cells. This deficiency not only restricts the in-depth study of the molecular mechanism of the disease, but also makes it difficult to support the development of precise treatment strategies based on cell subpopulations. An innovative method is urgently needed to fill the above technical gaps. Summary of the invention

[0006] The embodiments of the present invention provide a method and system for intelligently analyzing endothelial cell heterogeneity in atherosclerosis, which can accurately analyze endothelial cell heterogeneity in atherosclerosis.

[0007] An embodiment of the present invention provides an intelligent analysis method for endothelial cell heterogeneity in atherosclerosis, comprising: Obtaining sample data of endothelial cells in atherosclerosis of the patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is obtained by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; Extracting features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; Dividing the endothelial cells into different subpopulations according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes used to distinguish the different subpopulations, and defining and characterizing the characteristics of each subpopulation by analyzing the expression pattern of the marker genes; By comparing the gene expression differences between different endothelial cell subpopulations, the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis are identified. The differentially expressed genes serve as key molecular markers for understanding changes in endothelial cell function and heterogeneity.

[0008] As an improvement of the above scheme, the endothelial cells are divided into different subgroups according to the gene expression characteristics of the endothelial cells and the similarity of gene expression to determine marker genes for distinguishing different subgroups, and the characteristics of each subgroup are defined and characterized by analyzing the expression pattern of the marker genes, including: Similarity measurement and probability calculation steps: determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; Low-dimensional vector layout optimization steps: in a low-dimensional space of preset dimensions, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the low-dimensional vector layout is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, and the number of iterations is set within a predetermined range and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; Clustering step based on density peak: after obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. For the case where there are no cells with higher density, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as clustering centers. The remaining cells are assigned to the clusters to which the nearest clustering centers belong, thereby realizing the division of endothelial cell subpopulations. Marker gene determination steps: For each subgroup after division, calculate the fold change of the average expression ratio of genes within the subgroup and genes in other subgroups, set a threshold to screen marker genes based on the ratio and the average expression of genes within the subgroup, and use gene set enrichment analysis method to perform biological function pathway enrichment analysis on marker genes for verification.

[0009] As an improvement of the above scheme, the gene expression differences between different endothelial cell subpopulations are compared to identify differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis. The differentially expressed genes serve as key molecular markers for understanding endothelial cell function changes and heterogeneity, including: Subgroup gene expression calculation substep: Based on the characteristics of marker genes of different subgroups, for the two selected endothelial cell subgroups, the number of cells in each subgroup is counted, and then the expression of genes in the corresponding subgroup is calculated, which serves as the basic data preparation for subsequent model construction and analysis; Model construction and parameter estimation operator steps based on negative binomial distribution: construct a negative binomial distribution model suitable for describing the distribution of gene expression. The model expression contains terms associated with cell characteristic variables, model parameters to be determined, and discrete parameters unique to negative binomial distribution; use the maximum likelihood estimation method to carry out model parameter estimation to provide a basis for subsequent estimation of gene expression; The sub-steps of constructing the support vector machine regression model are as follows: the training data is set to consist of cell feature vectors and gene expression quantities, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the related terms of the slack variables, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter; various parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem; Differential expression analysis and screening sub-step: using the gene expression data estimated by the negative binomial distribution model as the input information of the support vector machine regression model, using the support vector machine regression model to deeply learn and analyze the differential relationship of gene expression between cell subpopulations; calculating the differential expression fold of genes between different subpopulations based on the adjusted expression data output by the support vector machine regression model; Test sub-step: Use a test method based on normal distribution approximation to perform a significance test on the calculated differential expression multiples, and screen out genes with significant differential expression based on a preset significance level threshold.

[0010] As an improvement of the above scheme, after comparing the gene expression differences between different endothelial cell subpopulations and identifying the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis, and the differentially expressed genes are used as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells, the method further includes: The protein products corresponding to the significantly differentially expressed genes were selected as potential drug targets, and the screened potential drug targets were verified by simulated functions.

[0011] As an improvement of the above scheme, the protein products corresponding to the significantly differentially expressed genes are selected as potential drug targets, and the screened potential drug targets are simulated for functional verification, including: Differentially expressed gene screening and sorting sub-step: For the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene are calculated by statistical test methods, and a comprehensive scoring function is constructed in combination with the gene function importance index. The function importance index is obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes are sorted in descending order according to the comprehensive scoring function; Protein product analysis and initial target screening sub-steps: Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting, locate multiple screening factors based on the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and localization characteristics of subcellular membrane proteins, and construct a screening function that includes the number of screening factors, the weight of each factor, and the protein's scoring function for each factor. Through this screening function combined with the set threshold, a candidate set of potential drug targets is screened out, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function determines the value according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; Binding affinity calculation sub-steps based on molecular docking: for each potential drug target, select candidate ligand molecules from a specific molecular library, first perform structural preprocessing on the target protein and the ligand molecule, then perform conformational search and docking in the active pocket region of the target protein, and calculate the binding energy using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. Determine partial charges and van der Waals interaction parameters based on the type of atomic chemical elements, set electrostatic interaction adjustment parameters based on experience, and set angle range parameters based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, select ligand-target combinations with strong binding affinity to preliminarily verify the drugability of the target; Sub-steps of target stability analysis based on dynamic simulation: For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; The target regulatory relationship verification sub-step based on network analysis is as follows: a gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and have weighted values. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of the index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

[0012] Another embodiment of the present invention provides an intelligent analysis system for endothelial cell heterogeneity in atherosclerosis, comprising: An acquisition module is used to acquire sample data of endothelial cells in atherosclerosis of a patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is acquired by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; A feature extraction module, used to extract features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; A subgroup division module, for dividing the endothelial cells into different subgroups according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes for distinguishing different subgroups, and define and characterize the characteristics of each subgroup by analyzing the expression pattern of the marker genes; The differential analysis module is used to compare the gene expression differences between different endothelial cell subpopulations and identify the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis. The differentially expressed genes serve as key molecular markers for understanding the changes in endothelial cell function and heterogeneity.

[0013] As an improvement of the above solution, the subgroup division module is specifically used for: Determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; In a low-dimensional space of a preset dimension, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the layout of the low-dimensional vector is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, the number of iterations is set within a predetermined range, and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; After obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. In the case of no higher density cells, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as cluster centers. The remaining cells are assigned to the cluster to which the nearest cluster center belongs, thereby realizing the division of endothelial cell subpopulations. For each subgroup after division, the fold change of the average expression ratio of genes within the subgroup to genes in other subgroups was calculated. A threshold was set based on the ratio and the average expression of genes within the subgroup to screen marker genes, and gene set enrichment analysis was used to verify the biological function pathway enrichment analysis of marker genes.

[0014] As an improvement of the above solution, the difference analysis module is specifically used for: Based on the characteristics of marker genes of different subpopulations, the number of cells in each subpopulation was counted for the two selected endothelial cell subpopulations, and then the expression of genes in the corresponding subpopulations was calculated, which served as the basic data preparation for subsequent model construction and analysis; A negative binomial distribution model suitable for describing the distribution of gene expression was constructed. The model expression contained terms associated with cell characteristic variables, as well as model parameters to be determined and discrete parameters unique to the negative binomial distribution. The maximum likelihood estimation method was used to estimate the model parameters, providing a basis for the subsequent estimation of gene expression. The training data is set to consist of cell feature vectors and gene expression, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the slack variable related terms, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter. The parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem. The gene expression data estimated by the negative binomial distribution model is used as the input information of the support vector machine regression model, and the support vector machine regression model is used to deeply learn and analyze the difference relationship of gene expression between cell subpopulations; the differential expression fold of genes between different subpopulations is calculated based on the adjusted expression data output by the support vector machine regression model; The calculated differential expression folds were tested for significance using a test method based on normal distribution approximation, and genes with significant differential expression were screened out based on the preset significance level threshold.

[0015] As an improvement of the above solution, the system further includes: The drug target verification module is used to select the protein products corresponding to the significantly differentially expressed genes as potential drug targets, and to perform simulated functional verification on the screened potential drug targets.

[0016] As an improvement of the above scheme, for the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene were calculated by statistical test methods, and a comprehensive scoring function was constructed in combination with the gene function importance index. The function importance index was obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes were sorted in descending order according to the comprehensive scoring function. Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting. According to the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and subcellular localization characteristics such as membrane protein localization, multiple screening factors are selected to construct a screening function including the number of screening factors, the weight of each factor, and the protein scoring function for each factor. The screening function is combined with the set threshold to screen out the candidate set of potential drug targets, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function is determined according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; For each potential drug target, candidate ligand molecules are selected from a specific molecular library. The target protein and ligand molecules are first pre-processed for structure, and then conformational search and docking are performed in the active pocket region of the target protein. The binding energy is calculated using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. The partial charge and van der Waals interaction parameters are determined based on the type of atomic chemical elements, the electrostatic interaction adjustment parameters are set based on experience, and the angle range parameters are set based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, ligand-target combinations with strong binding affinity are screened out to preliminarily verify the drugability of the target. For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; A gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and are weighted. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of this index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

[0017] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: Based on single-cell sequencing technology, the gene expression information of endothelial cells of patients with atherosclerosis is obtained. By extracting and analyzing the features of these original sequencing data, the endothelial cells are divided into different subgroups according to the similarity of gene expression, and the marker genes and their expression patterns of each subgroup are determined. Subsequently, by comparing the differences in gene expression between different subgroups, the differentially expressed genes that are significantly upregulated or downregulated in the process of atherosclerosis are identified as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells. Compared with the existing technology, which cannot accurately reveal the complex heterogeneity and dynamic changes inside endothelial cells during atherosclerosis, the embodiments of the present invention can accurately analyze the heterogeneity of endothelial cells through in-depth analysis at single-cell resolution, thereby revealing the functional changes and potential molecular markers at the level of individual cells, and ultimately achieving a comprehensive understanding of the molecular mechanism of the disease and discovering new therapeutic targets. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of a process of an intelligent analysis method of endothelial cell heterogeneity in atherosclerosis provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of the structure of an intelligent analysis system for endothelial cell heterogeneity in atherosclerosis provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] See also Figure 1 , is a flow chart of an intelligent analysis method for endothelial cell heterogeneity in atherosclerosis provided by an embodiment of the present invention. The intelligent analysis method for endothelial cell heterogeneity in atherosclerosis includes steps S10 to S13: S10, obtaining sample data of endothelial cells in atherosclerosis of the patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is obtained by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; S11, extracting features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; S12, dividing the endothelial cells into different subpopulations according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine marker genes for distinguishing different subpopulations, and defining and characterizing the characteristics of each subpopulation by analyzing the expression pattern of the marker genes; S13, compare the gene expression differences between different endothelial cell subpopulations, identify the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis, and use the differentially expressed genes as key molecular markers for understanding endothelial cell functional changes and heterogeneity.

[0021] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: Based on single-cell sequencing technology, the gene expression information of endothelial cells of patients with atherosclerosis is obtained. By extracting and analyzing the features of these original sequencing data, the endothelial cells are divided into different subgroups according to the similarity of gene expression, and the marker genes and their expression patterns of each subgroup are determined. Subsequently, by comparing the differences in gene expression between different subgroups, the differentially expressed genes that are significantly upregulated or downregulated in the process of atherosclerosis are identified as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells. Compared with the existing technology, which cannot accurately reveal the complex heterogeneity and dynamic changes inside endothelial cells during atherosclerosis, the embodiments of the present invention can accurately analyze the heterogeneity of endothelial cells through in-depth analysis at single-cell resolution, thereby revealing the functional changes and potential molecular markers at the level of individual cells, and ultimately achieving a comprehensive understanding of the molecular mechanism of the disease and discovering new therapeutic targets.

[0022] As an example, in step S10, the sample acquisition and processing process is as follows: Clinical sample collection: Patients clinically diagnosed with atherosclerosis were selected. During vascular surgery (such as carotid endarterectomy or coronary artery bypass grafting), experienced surgeons used sterile, pyrogen-free fine surgical instruments to accurately cut small pieces of diseased arterial tissue and adjacent relatively normal tissue. The volume of each sample was about 3-8 cubic millimeters. After the sample was collected, it was immediately placed in pre-cooled saline containing antibiotics (penicillin 100U / mL and streptomycin 100μg / mL) and antifungal agents (amphotericin B 0.25μg / mL), and quickly transported to the cell processing laboratory on ice. The whole process was controlled within 20 minutes.

[0023] Enzymatic hydrolysis and cell separation: A special complex enzyme solution consisting of 0.12% collagenase I, 0.08% trypsin and 0.03% hyaluronidase was used. The arterial tissue sample was placed in the enzyme solution and gently oscillated at 80-100 rpm in a constant temperature water bath oscillator at 37°C, saturated humidity and 5% CO2 for 25-35 minutes.

[0024] After the enzymatic hydrolysis was completed, an equal volume of DMEM / F12 medium containing 15% fetal bovine serum was added to terminate the enzyme reaction, and then filtered through a cell filter with a pore size of 30 μm to collect the filtrate. The filtrate was centrifuged at 400×g for 8 minutes, the supernatant was discarded, and the precipitate was washed twice with pre-cooled PBS buffer supplemented with 2mM EDTA. Finally, the cells were resuspended in an appropriate amount of single cell isolation buffer containing 0.04% BSA to adjust the cell concentration to about 8×10 5 -1.2×10 6 / ml to obtain high-quality single-cell suspension.

[0025] Using a single-cell capture system based on laser-induced fluorescence detection technology, and using single-cell analysis technology (scRNA-seq), single-cell suspensions are mixed with reverse transcription primers with unique molecular identifiers (UMIs), fluorescent labels, and reaction reagents. In the microchannels of the microfluidic chip, single cells are identified by laser-induced fluorescence detection, and combined with reverse transcription primers, and then encapsulated by oil droplets. In the oil droplets, the cells are lysed, and the released mRNA hybridizes with the reverse transcription primers and undergoes a reverse transcription reaction to generate cDNA with barcodes and fluorescent labels.

[0026] After collecting the oil droplets and breaking the emulsion, the microbeads with cDNA are screened and enriched according to the fluorescent label using fluorescence activated cell sorting (FACS) technology to remove impurities and microbeads that have not reacted successfully. The cDNA is then purified and amplified. Using a specially designed ultra-high precision library preparation kit, the amplified cDNA is first fragmented, and the fragment length is controlled at 220-480bp. Then, end repair, A-tail addition, adapter connection, and PCR amplification are performed in sequence to finally construct a library suitable for ultra-deep high-throughput sequencing. Sequencing is performed using the Illumina HiSeqXTen sequencing platform, and the double-end sequencing mode with a sequencing read length of 180-220bp is set to ensure that sufficient gene expression information is obtained.

[0027] Data cleaning and filtering: Using a self-written data processing script, we first removed the reads with a base quality value lower than 25 accounting for more than 15% of the sequencing data, and the reads containing ambiguous bases (N) greater than 5%. To address the problem of adapter sequence contamination, we accurately compared it with a known adapter sequence database to remove the matching adapter sequence part in the read. Genes with extremely low gene expression (total UMI counts in all cells less than 8) were removed to reduce noise interference in the data.

[0028] As an improvement of the above embodiment, the endothelial cells are divided into different subgroups according to the gene expression characteristics of the endothelial cells and the similarity of gene expression to determine marker genes for distinguishing different subgroups, and the characteristics of each subgroup are defined and characterized by analyzing the expression pattern of the marker genes, including: Similarity measurement and probability calculation steps: determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; Low-dimensional vector layout optimization steps: in a low-dimensional space of preset dimensions, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the low-dimensional vector layout is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, and the number of iterations is set within a predetermined range and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; Clustering step based on density peak: after obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. For the case where there are no cells with higher density, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as clustering centers. The remaining cells are assigned to the clusters to which the nearest clustering centers belong, thereby realizing the division of endothelial cell subpopulations. Marker gene determination steps: For each subgroup after division, calculate the fold change of the average expression ratio of genes within the subgroup and genes in other subgroups, set a threshold to screen marker genes based on the ratio and the average expression of genes within the subgroup, and use gene set enrichment analysis method to perform biological function pathway enrichment analysis on marker genes for verification.

[0029] In this embodiment, around the intelligent analysis of endothelial cell heterogeneity in atherosclerosis, first through the similarity measurement and probability calculation steps, according to the endothelial cell gene expression characteristics in high-dimensional space to construct cell vectors, calculate the Gaussian kernel similarity and corresponding probability, for subsequent analysis to lay the foundation, aiming to accurately capture the relationship between cells. Then in the low-dimensional vector layout optimization step, the corresponding vector is set in the preset low-dimensional space, and the reasonable layout is determined by minimizing the objective function and iterative optimization to ensure that the similarity structure of the high-dimensional space can be effectively restored. Subsequently, a clustering algorithm based on density peaks is used to calculate the local density and relative distance of cells based on the optimized low-dimensional vectors, and the clustering center is selected accordingly to achieve endothelial cell subpopulation division. Finally, in the marker gene determination step, the average expression of each subgroup gene is compared, and a threshold is set for screening and the marker gene is verified by gene set enrichment analysis, so as to accurately characterize the characteristics of each subgroup. This embodiment can realize accurate subgroup division of endothelial cells in atherosclerosis based on gene expression similarity, effectively screen out reliable marker genes that can distinguish subgroups, overcome the shortcomings of traditional analysis methods in analyzing endothelial cell heterogeneity, and provide strong data support for in-depth exploration of the disease mechanism of atherosclerosis and subsequent related research.

[0030] For ease of understanding, the working process of this embodiment is specifically described as follows: 1. Similarity measurement and probability calculation steps: Assume that the endothelial cell gene expression feature matrix is , the dimension is ,in is the total number of cells, i.e. the total number of endothelial cells studied, is the gene expression feature dimension, which represents the number of different aspects describing the gene expression characteristics of cells. and , whose vectors in the high-dimensional gene expression feature space are recorded as and ( , that is, each vector has dimensions, and the values ​​of these dimensions correspond to the specific values ​​of the cells in terms of various gene expression characteristics).

[0031] Calculate Gaussian kernel similarity between cells , the formula is ,here , indicating cells and The Euclidean distance in high-dimensional space measures the difference between the gene expression feature vectors of the two. and Cells and The gene expression feature vector The value of the dimension; is the scale parameter, ( is the cell vector mean, ), The similarity measurement range is determined based on the gene expression distribution of the cell population.

[0032] based on Calculating conditional probabilities , reflecting known cell Time Cell The probability of occurrence is obtained by normalizing the Gaussian kernel similarity between the cell and other cells. Represents cells With cells The Gaussian kernel similarity between them. Then calculate the joint probability : ( is the total number of cells, and Similarly, the conditional probability is calculated based on the same Gaussian kernel similarity, and the calculation formula is: , which reflects the known cell In the case of cells probability of occurrence), comprehensively describing the relationship between cells.

[0033] 2. Low-dimensional vector layout optimization steps: Set the low-dimensional space dimension , for high-dimensional cell vectors In the low-dimensional space, let the corresponding low-dimensional vector ( ), randomly initialized In a specific range (such as ) are evenly distributed.

[0034] Constructing the objective function : ,in is the expected cell distribution probability, and the calculation formula is , used with Comparison measures the difference between actual and expected cell distribution; , and the cells in the low-dimensional space and The vectors of are related, reflecting the influence of their distance relationship in low-dimensional space; . The first term measures the difference between the actual and expected cell distributions, and the second term is the regularization term that controls the smoothness of the low-dimensional space. (such as 0.5) is the regularization parameter to adjust its influence. (such as 0.1), (such as 0.2) Impact Then act on the objective function; Represents a low-dimensional vector No. The value of the dimension; Represents a low-dimensional vector No. The value of the dimension; Represents a low-dimensional vector No. The value of the dimension, is the vector corresponding to a high-dimensional cell A vector in a low-dimensional space, To identify cells, Used to identify the dimension sequence number.

[0035] Iterative optimization using gradient descent , each iteration , calculate the gradient Post Update : , is a low-dimensional vector In the The value at the iteration; is a low-dimensional vector In the The value at the iteration; (Initially set to 0.02) is the learning rate, and according to certain rules (such as every 5 iterations, ) decay, the number of iterations is set to 80-100 times, and the convergence threshold is set to .

[0036] 3. Clustering steps based on density peak: With optimized Calculation of local cell density : ,in Construct a distance matrix for the Euclidean distance between cells in low-dimensional space , ( yes The mean of is the standard deviation, is a low-dimensional space dimension), Reflection Cell The degree of surrounding aggregation.

[0037] Counting cells Minimum distance to higher density cells : , without higher density cells, Set as Maximum value. and Build Decision diagram (constructed using existing decision diagram technology), select and The large cells are clustered as centers (with is the horizontal axis, For the vertical axis, select and The cells with larger average values ​​are taken as cluster centers, and the remaining cells are assigned to corresponding clusters according to their distance from the cluster centers to achieve subgroup division.

[0038] 4. Marker gene determination steps: For the subgroups after division, set subgroup ZhongGene Average expression ( For gene In cells Normalized expression level, For subgroup The average expression level of other subpopulations was calculated by the same method. The change factor of the average expression ratio was calculated. ( Another subgroup ZhongGene The average expression level of To measure genes In different subgroups (such as and )), according to and the average gene expression level within the subgroup to set the threshold (e.g. The marker genes were screened by using gene set enrichment analysis (GSEA) to verify their enrichment in biological functional pathways. For specific gene set enrichment analysis, please refer to the existing gene set enrichment analysis method (GSEA).

[0039] As an improvement of the above embodiment, the gene expression differences between different endothelial cell subpopulations are compared to identify differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis, and the differentially expressed genes are used as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells, including: Subgroup gene expression calculation substep: Based on the characteristics of marker genes of different subgroups, for the two selected endothelial cell subgroups, the number of cells in each subgroup is counted, and then the expression of genes in the corresponding subgroup is calculated, which serves as the basic data preparation for subsequent model construction and analysis; Model construction and parameter estimation operator steps based on negative binomial distribution: construct a negative binomial distribution model suitable for describing the distribution of gene expression. The model expression contains terms associated with cell characteristic variables, model parameters to be determined, and discrete parameters unique to negative binomial distribution; use the maximum likelihood estimation method to carry out model parameter estimation to provide a basis for subsequent estimation of gene expression; The sub-steps of constructing the support vector machine regression model are as follows: the training data is set to consist of cell feature vectors and gene expression quantities, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the related terms of the slack variables, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter; various parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem; Differential expression analysis and screening sub-step: using the gene expression data estimated by the negative binomial distribution model as the input information of the support vector machine regression model, using the support vector machine regression model to deeply learn and analyze the differential relationship of gene expression between cell subpopulations; calculating the differential expression fold of genes between different subpopulations based on the adjusted expression data output by the support vector machine regression model; Test sub-step: Use a test method based on normal distribution approximation to perform a significance test on the calculated differential expression multiples, and screen out genes with significant differential expression based on a preset significance level threshold.

[0040] In the present embodiment, first, through the sub-step of calculating the gene expression of the sub-group, the number of cells is counted according to the characteristics of the marker genes, and the average expression of the gene in each sub-group is calculated to provide basic data for the subsequent model construction. Then, in the model construction and parameter estimation operator step based on the negative binomial distribution, the model is constructed and the parameters are estimated according to the distribution characteristics of the gene expression, thereby laying the foundation for the accurate estimation of the gene expression. Subsequently, in the sub-step of building the support vector machine regression model, the model is constructed according to the specific training data, and the model parameters are determined by solving the quadratic programming problem to enhance the analysis ability of the gene expression data. In the subsequent differential expression analysis and screening sub-step, the gene expression data estimated by the negative binomial distribution model is input into the support vector machine regression model, the gene expression difference relationship between the cell sub-groups is analyzed, and the differential expression multiple is calculated. Finally, in the test sub-step, the difference expression multiple is tested for significance using the test method of normal distribution approximation, and the significantly differentially expressed genes are screened out, as the key molecular markers for understanding the changes in endothelial cell function and heterogeneity, providing key information for the research and diagnosis and treatment of atherosclerotic diseases. This embodiment overcomes the shortcomings of traditional methods in identifying differentially expressed genes in endothelial cell subpopulations, and can accurately identify differentially expressed genes with significant changes between endothelial cell subpopulations in atherosclerosis, providing data support for in-depth exploration of the intrinsic molecular mechanisms of atherosclerosis, developing new diagnostic methods, and optimizing treatment strategies.

[0041] For ease of understanding, the working process of this embodiment is specifically described as follows: 1. Sub-steps for calculating subpopulation gene expression: Assume that there are two endothelial cell subsets and , for genes ,use Indicates gene In subgroups The number of cells in a subpopulation This gene is included The total number of cells; Indicates gene In subgroups The number of cells in the In subgroups Middle cell The expression level is recorded as (This expression level has been standardized in the previous stage and can truly reflect the actual expression situation). In subgroups The average expression level in The calculation formula is: Similarly, genes In subgroups The average expression level in Through such calculations, we can clearly know the average expression levels of genes in different subgroups and prepare basic data for subsequent model application.

[0042] 2. Model construction and parameter estimation operator steps based on negative binomial distribution: Constructing a negative binomial distribution model ,here Representative gene In cells The expression level in . Indicates gene In cells The expected expression in is calculated as ,in It's cells Characteristic variables, such as the specific subpopulation of cells and cell-specific metabolic indicators, characterize cell characteristics from many aspects and affect the calculation of expected expression. is a model parameter, and different value combinations affect The calculation results determine how well the model fits the gene expression distribution. It is a discrete parameter unique to the negative binomial distribution and is used to reflect the gene The discrete characteristics of expression distribution. Different genes have different expression characteristics, so the parameter value is different.

[0043] Constructing the likelihood function To carry out parameter estimation, its expression is: ; In the formula, Represents the total number of cells involved in the analysis, covering all endothelial cells of interest. is a gamma function, which is used to process complex operations such as factorials and integrals, and to assist in constructing a likelihood function that conforms to the distribution characteristics of gene expression. Take the derivative and set it to 0, and solve for and The value of is used to determine the specific form of the negative binomial distribution model, providing a basis for subsequent gene expression estimation.

[0044] 3. Support vector machine regression model construction sub-steps: Set the training data to ,in For cells The feature vector integrates multiple information such as the subpopulation to which the cell belongs and indicators related to cell function. Its dimension depends on the number of selected cell features. The value of each dimension reflects the corresponding feature and provides input content for model analysis. Represents gene expression (similar to the above situation), is the target variable to be predicted and analyzed by the model.

[0045] Construct a support vector machine regression model, whose objective function Set to: ;exist middle, is the weight vector, whose dimension is the same as the cell feature vector Dimension association, the weight of each element corresponding to the different characteristics of the cell in the model, determines the degree of influence of each characteristic on the result; Represents the number of samples of training data, that is, the cell-gene expression data pairs used to train the support vector machine regression model The number of is the square of the norm of the weight vector, which controls the complexity of the model during model training to prevent overfitting. , is the weight vector The dimension, is the weight vector No. elements, corresponding to the cell feature vector Middle The weight of a feature in the model determines the influence of the feature on the model prediction results. It is a penalty parameter used to adjust the emphasis on training errors. The larger the value, the heavier the error penalty, which makes the model fit the training data more closely. and It is a slack variable, which is used to deal with the situation where some data points in the actual data are difficult to meet the constraints. The restrictions are appropriately relaxed to make the model adapt to the real data distribution.

[0046] Also set constraints: , . is the weight vector The transpose of .

[0047] Solve this quadratic programming problem based on the objective function and constraints through an existing suitable optimization algorithm to determine the parameters of the support vector machine regression model Etc., so that the model can accurately analyze gene expression based on the input cell feature vector and achieve effective mapping from input to output.

[0048] 4. Differential expression analysis and screening sub-steps: The gene expression data estimated by the negative binomial distribution model (i.e., calculated according to the previously determined model and parameters) The estimated values ​​(such as the estimated values) are used as the input features of the support vector machine regression model and input into the support vector machine regression model with determined parameters. Based on the input feature information, the model deeply mines and analyzes the difference in gene expression between cell subpopulations and outputs the adjusted expression data. ( Is the model for gene In cells The adjusted estimate of expression in the mitochondria is a fusion of the model’s comprehensive analysis of cell characteristics and gene expression).

[0049] For the two selected subgroups and , computational genetics Among different subgroups ( and ) , the calculation formula is , Indicates gene In subgroups The number of cells in a subpopulation This gene is included The total number of cells; Indicates gene In subgroups Through this calculation, the expression differences of genes in different subpopulations can be intuitively and quantitatively seen, providing key quantitative indicators for screening significantly differentially expressed genes and helping to accurately locate genes that have important indicative effects on endothelial cell function changes and heterogeneity.

[0050] 5. Inspection sub-steps: Based on the normal distribution approximation Fold differential expression Conduct a significance test. And the corresponding sample data (involving the gene expression data of each subgroup) to calculate the test statistic (according to The conventional mathematical formula of the test is obtained by calculating the sample mean, sample standard deviation and other related data). Then, The value is consistent with the preset significance level threshold (for example, set to 0.05, corresponding to the corresponding Distribution critical value) comparison, filter out Values ​​less than this threshold (such as ) genes, which are genes with significantly differential expression. They are of great significance for a deep understanding of the functional changes and heterogeneous characteristics of endothelial cells in the process of atherosclerosis. They can serve as key molecular markers and provide important references for subsequent disease mechanism research, diagnosis and treatment.

[0051] As an improvement of the above embodiment, after comparing the gene expression differences between different endothelial cell subpopulations, identifying the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis, and the differentially expressed genes are used as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells, the method further includes: The protein products corresponding to the significantly differentially expressed genes were selected as potential drug targets, and the screened potential drug targets were verified by simulated functions.

[0052] Specifically, the protein products corresponding to the significantly differentially expressed genes are selected as potential drug targets, and the screened potential drug targets are simulated and functionally verified, including: Differentially expressed gene screening and sorting sub-step: For the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene are calculated by statistical test methods, and a comprehensive scoring function is constructed in combination with the gene function importance index. The function importance index is obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes are sorted in descending order according to the comprehensive scoring function; Protein product analysis and initial target screening sub-steps: Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting, locate multiple screening factors based on the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and localization characteristics of subcellular membrane proteins, and construct a screening function that includes the number of screening factors, the weight of each factor, and the protein's scoring function for each factor. Through this screening function combined with the set threshold, a candidate set of potential drug targets is screened out, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function determines the value according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; Binding affinity calculation sub-steps based on molecular docking: for each potential drug target, select candidate ligand molecules from a specific molecular library, first perform structural preprocessing on the target protein and the ligand molecule, then perform conformational search and docking in the active pocket region of the target protein, and calculate the binding energy using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. Determine partial charges and van der Waals interaction parameters based on the type of atomic chemical elements, set electrostatic interaction adjustment parameters based on experience, and set angle range parameters based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, select ligand-target combinations with strong binding affinity to preliminarily verify the drugability of the target; Sub-steps of target stability analysis based on dynamic simulation: For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; The target regulatory relationship verification sub-step based on network analysis is as follows: a gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and have weighted values. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of the index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

[0053] In the present embodiment, after identifying the differentially expressed genes of endothelial cell subpopulations in the process of atherosclerosis, further excavation of potential targets that can be used for drug development. First, through the differentially expressed gene screening and sorting sub-step, the differentially expressed genes are prioritized by comprehensively considering the differential expression degree and functional importance of the genes to focus on key genes. Then in the protein product analysis and target initial screening sub-step, according to the various functional characteristics and positioning characteristics of the protein, the potential drug target candidates are preliminarily screened out using a carefully constructed screening function. Then, through the binding affinity calculation sub-step based on molecular docking, a complex molecular docking algorithm formula is used to evaluate the binding ability of the ligand and the target from the molecular level, screen out a combination with strong binding affinity, and preliminarily verify the drugability of the target. Afterwards, in the target stability analysis sub-step based on kinetic simulation, a simulation system is constructed and according to the kinetic simulation algorithm formula, by monitoring the target protein structure stability index, the stability of the target after binding with the ligand is further investigated to enhance the reliability of target verification. Finally, in the target regulation relationship verification sub-step based on network analysis, a gene regulation network is constructed and the network analysis algorithm formula is used to explore the criticality of the target from the gene regulation level, and the rationality of the target is verified in an all-round way. This embodiment can realize the efficient and accurate screening of targets with potential drug development value based on the gene differential expression data of atherosclerotic endothelial cells, and improve the accuracy and reliability of target screening through multi-dimensional and multi-level simulation verification methods, providing high-quality target information for subsequent drug development for atherosclerosis, and effectively promoting the transformation process from basic research to clinical drug development.

[0054] For ease of understanding, the working process of this embodiment is specifically described as follows: 1. Differentially expressed gene screening and sorting sub-steps: Assume that the obtained differentially expressed gene set is , for each gene , through specific statistical tests (such as fold change of differential expression and statistical test values At the same time, based on the criticality of genes involved in biological pathways (Gene The degree of criticality in biological pathways is a factor in assessing the importance of gene function. The larger the value, the more critical the gene is in the biological pathway. The degree of association with known disease-related genes (Gene The degree of association with known disease-related genes is used to measure the correlation between genes and diseases. The higher the value, the closer the connection with disease-related genes) and other factors are used to comprehensively evaluate the gene function importance index. (Gene Functional importance index of ,in and is a weight coefficient pre-set according to the importance of each factor (assuming ). Construction of genes Comprehensive scoring function ,in accordance with The differentially expressed genes were sorted in descending order to identify those genes with significant changes in expression levels and that may play a key role in function, providing a priority order for subsequent target screening.

[0055] 2. Protein product analysis and target initial screening steps: Use bioinformatics tools such as UniProt to obtain the protein product set corresponding to the differentially expressed genes after sorting Based on the protein's involvement in cell signaling (protein Involvement in cell signaling is a factor in screening potential drug targets; A collection of protein products The protein products), and lipid metabolism (protein Interaction with lipid metabolism, used to evaluate the association between protein and lipid metabolism) and inflammatory factors (protein The interaction with inflammatory factors, reflecting the relationship between proteins and the inflammatory process, and the localization characteristics of membrane proteins (protein The localization characteristics of membrane proteins, such as whether they are transmembrane proteins, are an important basis for screening) and other screening factors to construct a screening function ,in is the number of screening factors (here ), It is The weight of each screening factor (e.g. Corresponding signal transduction, Corresponding to lipid metabolism, Corresponding to inflammatory factors, Corresponding to membrane protein localization), It's protein About Scoring function for each screening factor (e.g. for membrane protein localization, if it is a transmembrane protein, ,otherwise ). Set the threshold (like ),choose The protein products of In this way, proteins with potential drug target characteristics are preliminarily screened out.

[0056] 3. Binding affinity calculation sub-steps based on molecular docking: Targeting each potential drug target , select a series of candidate ligand molecules from a library of known drug molecules or a library of virtually designed small molecule compounds For target proteins and ligand molecules , first perform structural preprocessing, including adding hydrogen atoms, optimizing charge distribution and other operations. Then perform conformational search and docking in the active pocket area of ​​the target protein, and use the following molecular docking algorithm formula to calculate the binding energy : ;in, is the binding energy between the ligand and the target protein, is the number of ligand atoms, is the number of target protein atoms, and Ligand atoms and protein atoms The partial charge of the atom is determined by quantum chemical calculations or empirical databases depending on the chemical element of the atom (e.g. for carbon atoms, The value may fluctuate within a certain range depending on the chemical environment in which it is located). It's an atom and atoms The distance between and is the van der Waals interaction parameter (depending on the atom type, with specific empirical values ​​for different atom pairs), is the electrostatic interaction tuning parameter (empirically set to 0.5 to balance the contributions of van der Waals and electrostatic interactions), It's an atom and atoms The angle between the chemical bond vectors and the geometric center vector connecting them, is a parameter that controls the range of the angle-related term (set to 0.3 nm based on the size and shape characteristics of the protein active pocket). By calculating the binding energy of different ligands to the target, the ligand with lower binding energy (such as kcal / mol) of ligand-target combinations, indicating that they have strong binding affinity, preliminarily verifying the drugability of the target.

[0057] 4. Target stability analysis sub-steps based on kinetic simulation: For the above ligand-target combinations with good binding energy, a simulation system including the ligand, target protein and surrounding solvent molecules (such as water molecules) is constructed. Atomic mass parameters are determined based on the chemical composition of atoms. (Based on the relative atomic mass of atoms in the periodic table and with appropriate corrections considering the isotope distribution), calculate the atomic mass according to the parameter settings of the molecular mechanics force field (such as AMBER, CHARMM, etc.) Potential Energy Function (including bond energy, angular energy, dihedral energy, van der Waals energy and electrostatic energy, etc., each energy term is accurately calculated based on the force field parameters and the geometric relationship between atoms) and the interaction force between atoms (determined by the calculation formula of the interaction between atoms in the force field, taking into account factors such as the distance between atoms, charge, and chemical bond type), according to the simulated temperature (such as 300 ) and pressure (such as 1 atmosphere) conditions determine the external force (Calculated by the ideal gas state equation and thermodynamic principles, which is determined according to the temperature and pressure conditions of the simulated environment. For example, when simulating a physiological environment, temperature and pressure will have a certain effect on molecules, thereby affecting the structure and dynamic behavior of molecules.) Use the molecular dynamics simulation algorithm formula for simulation: ; Set the simulation time to nanoseconds, with a time step of During the simulation, the structural stability indicators of the target protein are monitored, such as the root mean square deviation (RMSD): (in It's an atom In time location, is the initial position). If during the simulation, the RMSD of the target protein remains within a small range (e.g. nanometers), indicating that the target protein has a stable structure after binding with the ligand, further verifying the effectiveness of the target.

[0058] 5. Sub-steps of target regulation relationship verification based on network analysis: Construct a gene regulatory network that includes differentially expressed genes, potential drug targets, and other related genes. The nodes in the network are genes and targets, and the edges represent the regulatory relationships between genes or between genes and targets (such as the binding relationship between transcription factors and target genes, the interaction between upstream and downstream molecules in signaling pathways, etc.). The weight of the edge can be assigned based on the credibility of experimental evidence or bioinformatics predictions. For each potential drug target , determine the gene set that has a direct or indirect regulatory relationship with the target based on the network topology and the shortest path length between gene and target (Calculated by the shortest path algorithm in graph theory, such as Dijkstra algorithm). Use the network analysis algorithm formula to calculate the importance index of the target in the gene regulatory network. : ,in and As defined in the previous differentially expressed gene screening. The comparison with the preset threshold (such as 0.8) determines the criticality of the target in the gene regulatory network. Higher, indicating the target It occupies a key position in the gene regulatory network and has a close regulatory relationship with multiple important differentially expressed genes. The rationality and importance of this target are verified from the gene regulatory level. Based on the above simulation function verification results, the most promising drug target is determined for subsequent drug research and development.

[0059] See also Figure 2 , is a schematic diagram of a structure of an intelligent analysis device for endothelial cell heterogeneity in atherosclerosis provided by an embodiment of the present invention. The intelligent analysis system for endothelial cell heterogeneity in atherosclerosis includes: The acquisition module 10 is used to obtain sample data of endothelial cells in atherosclerosis of the patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is obtained by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; A feature extraction module 11 is used to extract features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; A subgroup division module 12 is used to divide the endothelial cells into different subgroups according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes used to distinguish different subgroups, and to define and characterize the characteristics of each subgroup by analyzing the expression pattern of the marker genes; The differential analysis module 13 is used to compare the gene expression differences between different endothelial cell subpopulations, and identify the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis. The differentially expressed genes serve as key molecular markers for understanding the changes in endothelial cell function and heterogeneity.

[0060] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: Based on single-cell sequencing technology, the gene expression information of endothelial cells of patients with atherosclerosis is obtained. By extracting and analyzing the features of these original sequencing data, the endothelial cells are divided into different subgroups according to the similarity of gene expression, and the marker genes and their expression patterns of each subgroup are determined. Subsequently, by comparing the differences in gene expression between different subgroups, the differentially expressed genes that are significantly upregulated or downregulated in the process of atherosclerosis are identified as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells. Compared with the existing technology, which cannot accurately reveal the complex heterogeneity and dynamic changes inside endothelial cells during atherosclerosis, the embodiments of the present invention can accurately analyze the heterogeneity of endothelial cells through in-depth analysis at single-cell resolution, thereby revealing the functional changes and potential molecular markers at the level of individual cells, and ultimately achieving a comprehensive understanding of the molecular mechanism of the disease and discovering new therapeutic targets.

[0061] As an improvement of the above solution, the subgroup division module is specifically used for: Determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; In a low-dimensional space of a preset dimension, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the layout of the low-dimensional vector is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, the number of iterations is set within a predetermined range, and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; After obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. In the case of no higher density cells, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as cluster centers. The remaining cells are assigned to the cluster to which the nearest cluster center belongs, thereby realizing the division of endothelial cell subpopulations. For each subgroup after division, the fold change of the average expression ratio of genes within the subgroup to genes in other subgroups was calculated. A threshold was set based on the ratio and the average expression of genes within the subgroup to screen marker genes, and gene set enrichment analysis was used to verify the biological function pathway enrichment analysis of marker genes.

[0062] As an improvement of the above solution, the difference analysis module is specifically used for: Based on the characteristics of marker genes of different subpopulations, the number of cells in each subpopulation was counted for the two selected endothelial cell subpopulations, and then the expression of genes in the corresponding subpopulations was calculated, which served as the basic data preparation for subsequent model construction and analysis; A negative binomial distribution model suitable for describing the distribution of gene expression was constructed. The model expression contained terms associated with cell characteristic variables, as well as model parameters to be determined and discrete parameters unique to the negative binomial distribution. The maximum likelihood estimation method was used to estimate the model parameters, providing a basis for the subsequent estimation of gene expression. The training data is set to consist of cell feature vectors and gene expression, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the slack variable related terms, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter. The parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem. The gene expression data estimated by the negative binomial distribution model is used as the input information of the support vector machine regression model, and the support vector machine regression model is used to deeply learn and analyze the difference relationship of gene expression between cell subpopulations; the differential expression fold of genes between different subpopulations is calculated based on the adjusted expression data output by the support vector machine regression model; The calculated differential expression folds were tested for significance using a test method based on normal distribution approximation, and genes with significant differential expression were screened out based on the preset significance level threshold.

[0063] As an improvement of the above solution, the system further includes: The drug target verification module is used to select the protein products corresponding to the significantly differentially expressed genes as potential drug targets, and to perform simulated functional verification on the screened potential drug targets.

[0064] As an improvement of the above scheme, for the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene were calculated by statistical test methods, and a comprehensive scoring function was constructed in combination with the gene function importance index. The function importance index was obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes were sorted in descending order according to the comprehensive scoring function. Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting. According to the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and subcellular localization characteristics such as membrane protein localization, multiple screening factors are selected to construct a screening function including the number of screening factors, the weight of each factor, and the protein scoring function for each factor. The screening function is combined with the set threshold to screen out the candidate set of potential drug targets, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function is determined according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; For each potential drug target, candidate ligand molecules are selected from a specific molecular library. The target protein and ligand molecules are first pre-processed for structure, and then conformational search and docking are performed in the active pocket region of the target protein. The binding energy is calculated using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. The partial charge and van der Waals interaction parameters are determined based on the type of atomic chemical elements, the electrostatic interaction adjustment parameters are set based on experience, and the angle range parameters are set based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, ligand-target combinations with strong binding affinity are screened out to preliminarily verify the drugability of the target. For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; A gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and are weighted. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of this index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

[0065] It should be noted that the system embodiment described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the system embodiment provided by the present invention, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines. A person of ordinary skill in the art may understand and implement it without paying any creative effort.

[0066] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. An intelligent analysis method for endothelial cell heterogeneity in atherosclerosis, characterized in that: include: Obtaining sample data of endothelial cells in atherosclerosis of the patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is obtained by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; Extracting features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; Dividing the endothelial cells into different subpopulations according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes used to distinguish the different subpopulations, and defining and characterizing the characteristics of each subpopulation by analyzing the expression pattern of the marker genes; By comparing the gene expression differences between different endothelial cell subpopulations, the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis are identified. The differentially expressed genes serve as key molecular markers for understanding changes in endothelial cell function and heterogeneity.

2. The intelligent analysis method of endothelial cell heterogeneity in atherosclerosis according to claim 1, characterized in that: The method of dividing the endothelial cells into different subgroups based on the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes used to distinguish the different subgroups, and defining and characterizing the characteristics of each subgroup by analyzing the expression pattern of the marker genes, includes: Similarity measurement and probability calculation steps: determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; Low-dimensional vector layout optimization steps: in a low-dimensional space of preset dimensions, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the low-dimensional vector layout is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, and the number of iterations is set within a predetermined range and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; Clustering step based on density peak: after obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. For the case where there are no cells with higher density, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as clustering centers. The remaining cells are assigned to the clusters to which the nearest clustering centers belong, thereby realizing the division of endothelial cell subpopulations. Marker gene determination steps: For each subgroup after division, calculate the fold change of the average expression ratio of genes within the subgroup and genes in other subgroups, set a threshold to screen marker genes based on the ratio and the average expression of genes within the subgroup, and use gene set enrichment analysis method to perform biological function pathway enrichment analysis on marker genes for verification.

3. The intelligent analysis method of endothelial cell heterogeneity in atherosclerosis according to claim 2, characterized in that: The gene expression differences between different endothelial cell subpopulations are compared to identify differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis. The differentially expressed genes serve as key molecular markers for understanding endothelial cell function changes and heterogeneity, including: Subgroup gene expression calculation substep: Based on the characteristics of marker genes of different subgroups, for the two selected endothelial cell subgroups, the number of cells in each subgroup is counted, and then the expression of genes in the corresponding subgroup is calculated, which serves as the basic data preparation for subsequent model construction and analysis; Model construction and parameter estimation operator steps based on negative binomial distribution: construct a negative binomial distribution model suitable for describing the distribution of gene expression. The model expression contains terms associated with cell characteristic variables, model parameters to be determined, and discrete parameters unique to negative binomial distribution; use the maximum likelihood estimation method to carry out model parameter estimation to provide a basis for subsequent estimation of gene expression; The sub-steps of constructing the support vector machine regression model are as follows: the training data is set to consist of cell feature vectors and gene expression quantities, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the related terms of the slack variables, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter; various parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem; Differential expression analysis and screening sub-step: using the gene expression data estimated by the negative binomial distribution model as the input information of the support vector machine regression model, using the support vector machine regression model to deeply learn and analyze the differential relationship of gene expression between cell subpopulations; calculating the differential expression fold of genes between different subpopulations based on the adjusted expression data output by the support vector machine regression model; Test sub-step: Use a test method based on normal distribution approximation to perform a significance test on the calculated differential expression multiples, and screen out genes with significant differential expression based on a preset significance level threshold.

4. The intelligent analysis method of endothelial cell heterogeneity in atherosclerosis according to claim 3, characterized in that: After comparing the gene expression differences between different endothelial cell subpopulations and identifying the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis, wherein the differentially expressed genes serve as key molecular markers for understanding the functional changes and heterogeneity of endothelial cells, the method further comprises: The protein products corresponding to the significantly differentially expressed genes were selected as potential drug targets, and the screened potential drug targets were verified by simulated functions.

5. The intelligent analysis method of endothelial cell heterogeneity in atherosclerosis according to claim 4, characterized in that: The protein products corresponding to the significantly differentially expressed genes are selected as potential drug targets, and the screened potential drug targets are simulated and functionally verified, including: Differentially expressed gene screening and sorting sub-step: For the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene are calculated by statistical test methods, and a comprehensive scoring function is constructed in combination with the gene function importance index. The function importance index is obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes are sorted in descending order according to the comprehensive scoring function; Protein product analysis and initial target screening sub-steps: Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting, locate multiple screening factors based on the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and localization characteristics of subcellular membrane proteins, and construct a screening function that includes the number of screening factors, the weight of each factor, and the protein's scoring function for each factor. Through this screening function combined with the set threshold, a candidate set of potential drug targets is screened out, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function determines the value according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; Binding affinity calculation sub-steps based on molecular docking: for each potential drug target, select candidate ligand molecules from a specific molecular library, first perform structural preprocessing on the target protein and the ligand molecule, then perform conformational search and docking in the active pocket region of the target protein, and calculate the binding energy using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. Determine partial charges and van der Waals interaction parameters based on the type of atomic chemical elements, set electrostatic interaction adjustment parameters based on experience, and set angle range parameters based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, select ligand-target combinations with strong binding affinity to preliminarily verify the drugability of the target; Sub-steps of target stability analysis based on dynamic simulation: For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; The target regulatory relationship verification sub-step based on network analysis is as follows: a gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and have weighted values. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of the index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

6. An intelligent analysis system for endothelial cell heterogeneity in atherosclerosis, characterized in that: include: An acquisition module is used to acquire sample data of endothelial cells in atherosclerosis of a patient; the sample data includes original sequencing data of gene expression information of a single endothelial cell, and the original sequencing data is acquired by analyzing the transcriptome level of the endothelial cells in advance using single-cell sequencing technology; A feature extraction module, used to extract features from the sample data of the endothelial cells to obtain gene expression features of the endothelial cells; A subgroup division module, for dividing the endothelial cells into different subgroups according to the gene expression characteristics of the endothelial cells and according to the gene expression similarity, so as to determine the marker genes for distinguishing different subgroups, and define and characterize the characteristics of each subgroup by analyzing the expression pattern of the marker genes; The differential analysis module is used to compare the gene expression differences between different endothelial cell subpopulations and identify the differentially expressed genes that are significantly upregulated or downregulated in the relevant endothelial cell subpopulations during atherosclerosis. The differentially expressed genes serve as key molecular markers for understanding the changes in endothelial cell function and heterogeneity.

7. The intelligent analysis system for endothelial cell heterogeneity in atherosclerosis according to claim 6, characterized in that: The subgroup division module is specifically used for: Determine the vector of the cell in the high-dimensional gene expression feature space using the obtained gene expression features, calculate the Gaussian kernel similarity between cells, and calculate the conditional probability and joint probability based on the similarity; In a low-dimensional space of a preset dimension, a low-dimensional vector is set corresponding to a high-dimensional cell vector, and the layout of the low-dimensional vector is determined by minimizing a specific objective function; in the optimization process, the Gaussian kernel similarity matrix, the conditional probability matrix and the joint probability are first calculated according to the gene expression feature data, and then the low-dimensional vector is randomly initialized so that it is evenly distributed in the low-dimensional space and is in a specific numerical range, and then the objective function is iteratively optimized by the gradient descent method, and the gradient of the objective function with respect to the low-dimensional vector is calculated at each iteration and the low-dimensional vector is updated accordingly, wherein the learning rate is initially set to a specific value and decays according to a predetermined rule according to the change of the objective function during the iteration process, the number of iterations is set within a predetermined range, and whether convergence is determined based on the comparison between the change of the objective function and a preset convergence threshold; After obtaining the optimized low-dimensional vector, a clustering algorithm based on density peak is used in the low-dimensional space to calculate the local density of cells and the minimum distance to cells with higher density. The local density calculation is based on the Euclidean distance between cells and the statistical parameters of the distance matrix to determine the calculation scale. In the case of no higher density cells, the minimum distance is set to a predetermined larger value. A decision graph is constructed based on the local density and the minimum distance, and cells with predetermined characteristics are selected as cluster centers. The remaining cells are assigned to the cluster to which the nearest cluster center belongs, thereby realizing the division of endothelial cell subpopulations. For each subgroup after division, the fold change of the average expression ratio of genes within the subgroup to genes in other subgroups was calculated. A threshold was set based on the ratio and the average expression of genes within the subgroup to screen marker genes, and gene set enrichment analysis was used to verify the biological function pathway enrichment analysis of marker genes.

8. The intelligent analysis system for endothelial cell heterogeneity in atherosclerosis according to claim 7, characterized in that: The difference analysis module is specifically used for: Based on the characteristics of marker genes of different subpopulations, the number of cells in each subpopulation was counted for the two selected endothelial cell subpopulations, and then the expression of genes in the corresponding subpopulations was calculated, which served as the basic data preparation for subsequent model construction and analysis; A negative binomial distribution model suitable for describing the distribution of gene expression was constructed. The model expression contained terms associated with cell characteristic variables, as well as model parameters to be determined and discrete parameters unique to the negative binomial distribution. The maximum likelihood estimation method was used to estimate the model parameters, providing a basis for the subsequent estimation of gene expression. The training data is set to consist of cell feature vectors and gene expression, and a support vector machine regression model is constructed. The objective function of the support vector machine regression model comprehensively considers the relevant norm of the weight vector and the slack variable related terms, and sets a constraint system based on the cell feature vector, weight vector, bias term, slack variable and insensitive loss parameter. The parameters of the support vector machine regression model are determined by solving the corresponding quadratic programming problem. The gene expression data estimated by the negative binomial distribution model is used as the input information of the support vector machine regression model, and the support vector machine regression model is used to deeply learn and analyze the difference relationship of gene expression between cell subpopulations; the differential expression fold of genes between different subpopulations is calculated based on the adjusted expression data output by the support vector machine regression model; The calculated differential expression folds were tested for significance using a test method based on normal distribution approximation, and genes with significant differential expression were screened out based on the preset significance level threshold.

9. The intelligent analysis system for endothelial cell heterogeneity in atherosclerosis according to claim 8, characterized in that: Also includes: The drug target verification module is used to select the protein products corresponding to the significantly differentially expressed genes as potential drug targets, and to perform simulated functional verification on the screened potential drug targets.

10. The intelligent analysis system for endothelial cell heterogeneity in atherosclerosis according to claim 9, characterized in that: The drug target verification module is specifically used for: For the obtained differentially expressed gene set, the differential expression fold change and statistical test value of each gene were calculated by statistical test methods, and a comprehensive scoring function was constructed in combination with the gene function importance index. The function importance index was obtained based on a multi-factor comprehensive evaluation of the criticality of the gene's participation in the biological pathway and the degree of association with known disease-related genes. The differentially expressed genes were sorted in descending order according to the comprehensive scoring function. Use bioinformatics tools to obtain the protein product set information corresponding to the differentially expressed genes after sorting. According to the functional characteristics of the protein, such as participation in cell signal transduction, interaction with lipid metabolism and inflammatory factors, and subcellular localization characteristics such as membrane protein localization, multiple screening factors are selected to construct a screening function including the number of screening factors, the weight of each factor, and the protein scoring function for each factor. The screening function is combined with the set threshold to screen out the candidate set of potential drug targets, where the weight of each screening factor is set according to its relative importance in drug target screening, and the scoring function is determined according to the degree of match between the protein characteristics and the screening criteria, thereby achieving preliminary screening of protein products to determine potential drug target candidates; For each potential drug target, candidate ligand molecules are selected from a specific molecular library. The target protein and ligand molecules are first pre-processed for structure, and then conformational search and docking are performed in the active pocket region of the target protein. The binding energy is calculated using a molecular docking algorithm formula that includes the number of atoms in the ligand and target protein, partial charges of atoms, distances between atoms, van der Waals interaction parameters, electrostatic interaction adjustment parameters, the angle between the chemical bond vector and the geometric center vector between atoms, and control angle range parameters. The partial charge and van der Waals interaction parameters are determined based on the type of atomic chemical elements, the electrostatic interaction adjustment parameters are set based on experience, and the angle range parameters are set based on the characteristics of the protein active pocket. By comparing the binding energy with the preset threshold, ligand-target combinations with strong binding affinity are screened out to preliminarily verify the drugability of the target. For ligand-target combinations with good binding energy, a simulation system including ligand, target protein and solvent molecules is constructed, the atomic mass parameters are determined according to the atomic chemical composition, the atomic potential energy function and the atomic interaction force are calculated according to the molecular mechanics force field parameter settings, the external force is determined according to the simulated environmental temperature and pressure conditions, the molecular dynamics simulation algorithm formula is used for simulation, the simulation time and step length are set, the root mean square deviation of the target protein structure stability index is monitored, and the structural stability of the target protein after binding with the ligand is judged based on the comparison of the root mean square deviation with the preset range, so as to further verify the effectiveness of the target; A gene regulatory network including differentially expressed genes, potential drug targets and other related genes is constructed. The network nodes are genes and targets, and the edges represent regulatory relationships and are weighted. For each potential drug target, the set of genes that have a regulatory relationship with the target and the shortest path length between the gene and the target are determined based on the network topology. The importance index of the target in the gene regulatory network is calculated using the network analysis algorithm formula. The criticality of the target in the gene regulatory network is judged based on the comparison of this index with the preset threshold, thereby verifying the rationality of the target from the gene regulation level.

Images