Processing method and device of sepsis data, storage medium and electronic equipment
By combining weighted gene co-expression network analysis and machine learning algorithms, we screened out target core markers related to sepsis, solved the problem of accuracy in early diagnosis of sepsis in children, and achieved efficient early detection.
Patent Information
- Application Number
- CN202510961385.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-11
- Publication Date
- 2025-10-17
AI Technical Summary
Existing technologies cannot effectively achieve early diagnosis and accurate detection of sepsis in children due to insufficient sensitivity and specificity, and existing biomarker detection methods lack precision.
By obtaining gene expression data of sepsis and healthy control samples, performing data preprocessing and batch effect correction, and using a combination of weighted gene co-expression network analysis and preset machine learning algorithms, we screened out target core markers related to sepsis and constructed a diagnostic model for early diagnosis.
It improves the accuracy of sepsis detection in children, achieves early diagnosis and precise detection, reduces noise interference, and enhances the effectiveness of the diagnostic model.
Smart Images
Figure CN120809165A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the medical technology field, in particular to a sepsis data processing method and device, a storage medium and an electronic equipment. BACKGROUND
[0002] Pediatric sepsis (PS) is a highly lethal immune system response triggered by infection, which poses a significant threat to the newborn and young child population. Early diagnosis and timely treatment of sepsis directly affect the survival rate of children.
[0003] Currently, the diagnosis of pediatric sepsis in clinical practice still relies on clinical symptoms and the determination of some biomarkers, but the sensitivity and specificity are limited, and early diagnosis of sepsis has not yet been effectively achieved.
[0004] With the development of genomics and big data technology, researchers have attempted to find potential biomarkers for sepsis through gene expression profiling analysis, but the current detection method for potential biomarkers of sepsis lacks precision and cannot be quickly and effectively detected.
[0005] Therefore, how to achieve early diagnosis and accurate detection of pediatric sepsis detection is a problem that needs to be solved by the present application. SUMMARY
[0006] Therefore, the present application discloses a sepsis data processing method and device, a storage medium and an electronic equipment, which aims to achieve the purpose of early diagnosis and accurate detection of pediatric sepsis detection.
[0007] In order to achieve the above purpose, the disclosed technical solution is as follows:
[0008] The first aspect of the present application discloses a sepsis data processing method, comprising:
[0009] obtaining first gene expression data of sepsis and second gene expression data of healthy control samples;
[0010] data preprocessing and batch effect correction are performed on the first gene expression data and the second gene expression data;
[0011] a plurality of differentially expressed genes are screened out from the processed and corrected first gene expression data and the processed and corrected second gene expression data by differential expression analysis;
[0012] a plurality of key module genes related to sepsis and an initial core marker related to sepsis are screened out from the differentially expressed genes by weighted gene co-expression network analysis;
[0013] The plurality of differentially expressed genes and the plurality of key module genes are subjected to intersection analysis to obtain a plurality of intersection genes.
[0014] The target core marker is screened out from the initial core marker by combining a preset machine learning algorithm with the plurality of intersection genes.
[0015] Preferably, the data preprocessing of the first gene expression data and the second gene expression data includes:
[0016] The ComBat function of the sva package is called to calculate the mean expression level and the mean difference between batches of the first gene expression data and the second gene expression data in all samples.
[0017] The mean difference between batches is adjusted by the empirical Bayes method.
[0018] The adjustment parameter of the adjusted mean difference between batches is calculated.
[0019] The variance of the first gene expression data and the second gene expression data is shrunk.
[0020] The first gene expression data after variance shrinkage and the second gene expression data after variance shrinkage are adjusted by the adjustment parameter to obtain an expression value matrix, so as to complete the process of data preprocessing of the first gene expression data and the second gene expression data.
[0021] Preferably, the batch effect correction of the first gene expression data and the second gene expression data includes:
[0022] The function correction is performed on the first gene expression data and the second gene expression data.
[0023] The batch variation source of the first gene expression data after function correction and the batch variation source of the second gene expression data after function correction are eliminated, so as to complete the process of batch effect correction of the first gene expression data and the second gene expression data.
[0024] Preferably, the plurality of differentially expressed genes at least includes a plurality of up-regulated genes and a plurality of down-regulated genes, and the plurality of differentially expressed genes is screened out from the processed corrected first gene expression data and the processed corrected second gene expression data by the differential expression analysis method, including:
[0025] A design matrix representing sample grouping is constructed, wherein the design matrix representing sample grouping at least includes a sepsis sample grouping and a healthy control sample grouping.
[0026] setting a contrast matrix; wherein the contrast matrix describes the contrast relationship between each gene expression data;
[0027] performing linear model fitting on the processed corrected first gene expression data and the processed corrected second gene expression data to obtain a statistic value and a probability value of gene expression difference; wherein the probability value is the probability of sample result or extreme result under a preset condition;
[0028] filtering the plurality of up-regulated genes from the processed corrected first gene expression data through the sepsis sample grouping, the contrast matrix, the statistic value of gene expression difference and the probability value;
[0029] filtering the plurality of down-regulated genes from the processed corrected second gene expression data through the healthy control sample grouping, the contrast matrix, the statistic value of gene expression difference and the probability value.
[0030] Preferably, filtering the plurality of key module genes related to sepsis and the initial core markers related to sepsis from the differential expression genes through the weighted gene co-expression network analysis method and the differential expression genes, comprising:
[0031] setting a soft threshold value through a soft threshold function;
[0032] identifying a plurality of gene modules from the differential expression genes through hierarchical clustering analysis and dynamic tree cutting method; wherein the gene module is a group of genes whose expression profiles show consistent changes under given conditions;
[0033] calculating a first correlation of feature genes in the plurality of gene modules with the processed corrected first gene expression data and a second correlation of the feature genes in the plurality of gene modules with the processed corrected second gene expression data;
[0034] determining the plurality of key module genes related to sepsis according to the first correlation, the second correlation and the soft threshold value;
[0035] filtering genes meeting the conditions from the plurality of key module genes related to sepsis as the initial core markers related to sepsis of a preset number.
[0036] Preferably, filtering the target core markers from the initial core markers through the preset machine learning algorithm combination and the plurality of intersection genes, comprising:
[0037] filtering the target core markers from the initial core markers through the preset machine learning algorithm combination consisting of LASSO regression algorithm, SVM-RFE algorithm and random forest algorithm and the plurality of intersection genes.
[0038] Preferably, it further comprises:
[0039] A diagnostic model is constructed according to the expression amount of the target core marker, and the serum of a patient is diagnosed by the diagnostic model.
[0040] The second aspect of the present application discloses a sepsis data processing device, comprising:
[0041] An acquisition unit is configured to acquire first gene expression data of sepsis and second gene expression data of a healthy control sample;
[0042] A processing correction unit is configured to perform data preprocessing and batch effect correction on the first gene expression data and the second gene expression data;
[0043] A first screening unit is configured to screen a plurality of differentially expressed genes from the first gene expression data and the second gene expression data after processing correction by a differential expression analysis method;
[0044] A second screening unit is configured to screen a plurality of key module genes related to sepsis and an initial core marker related to sepsis from the differentially expressed genes by a weighted gene co-expression network analysis method;
[0045] An intersection analysis unit is configured to perform intersection analysis on the plurality of differentially expressed genes and the plurality of key module genes to obtain a plurality of intersection genes;
[0046] A third screening unit is configured to screen a target core marker from the initial core marker by a preset machine learning algorithm combination and the plurality of intersection genes.
[0047] The third aspect of the present application discloses a storage medium, comprising stored instructions, wherein the instructions control the device where the storage medium is located to execute the sepsis data processing method of any one of the first aspect when the instructions are executed.
[0048] The fourth aspect of the present application discloses an electronic device, comprising a memory and one or more instructions, wherein the one or more instructions are stored in the memory and are configured to be executed by one or more processors to execute the sepsis data processing method of any one of the first aspect.
[0049] The technical scheme can know that the application discloses a sepsis data processing method, device, storage medium and electronic equipment, first gene expression data of sepsis and second gene expression data of a healthy control sample are acquired, the first gene expression data and the second gene expression data are preprocessed and batch effect corrected, a plurality of differential expression genes are screened out from the first gene expression data and the second gene expression data after processing and correction through a differential expression analysis method, a plurality of key module genes related to sepsis and an initial core marker related to sepsis are screened out through a weighted gene co-expression network analysis method and the differential expression genes, intersection analysis is performed on the plurality of differential expression genes and the plurality of key module genes, a plurality of intersection genes are obtained, and a target core marker is screened out from the initial core marker through a preset machine learning algorithm combination and the plurality of intersection genes.
[0050] The application has the beneficial effects that the target core marker associated with sepsis is screened out through the weighted gene co-expression network analysis method and the preset machine learning algorithm combination, the genes with similar expression patterns are clustered into modules through the weighted gene co-expression network analysis, the modules are associated with sepsis, and the soft threshold optimization in the weighted gene co-expression network analysis makes the subsequent screening of the target core marker more consistent with the scale-free network distribution, reduces noise interference, improves the accuracy of obtaining the target core marker associated with sepsis, and a diagnosis model can be constructed according to the expression amount of the target core marker, and the diagnosis model is used to diagnose the serum of a patient, so that the early diagnosis and accurate detection of child sepsis detection are achieved.
[0051] The above description is only a summary of the technical scheme of the application, in order to more clearly understand the technical means of the application, the specific embodiments of the application can be implemented according to the content of the specification, and in order to make the above and other purposes, characteristics and advantages of the application more obvious and easy to understand, the following specific embodiments of the application are described. BRIEF DESCRIPTION OF DRAWINGS
[0052] Various other advantages and benefits will become apparent to those of ordinary skill in the art, upon reading the following detailed description of the preferred embodiments. The accompanying drawings are included to provide a description of preferred embodiments, and are not meant to limit the present application. Furthermore, the same reference numerals are used throughout the several drawings to represent similar components. In the drawings:
[0053] Figure 1 A flowchart of a sepsis data processing method disclosed by an embodiment of the application is shown in the figure;
[0054] Figure 2 A schematic diagram of differential expression genes disclosed by an embodiment of the application is shown in the figure;
[0055] Figure 3A schematic diagram of the weighted gene co-expression network analysis method disclosed in the embodiments of the present application;
[0056] Figure 4 A schematic diagram of the preset machine learning algorithm combination disclosed in the embodiments of the present application;
[0057] Figure 5 A schematic diagram of the protein expression level of the target core marker disclosed in the embodiments of the present application;
[0058] Figure 6 A structural schematic diagram of a sepsis data processing device disclosed in the embodiments of the present application;
[0059] Figure 7 A structural schematic diagram of an electronic device disclosed in the embodiments of the present application. DETAILED DESCRIPTION
[0060] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without any creative work fall within the scope of protection of the present application.
[0061] In the present application, the term “comprising” or “including” or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement “comprising a” does not exclude the presence of another identical element in the process, method, article or device including the element.
[0062] As known from the background, the current clinical diagnosis of pediatric sepsis still relies on clinical symptoms and determination of some biomarkers, but the sensitivity and specificity are limited, and early diagnosis of sepsis has not been effectively achieved. The detection method for potential biomarkers of sepsis lacks precision and cannot be quickly and effectively detected.
[0063] To solve the above problems, the application discloses a sepsis data processing method, device, storage medium and electronic equipment, which combines a weighted gene co-expression network analysis method with a preset machine learning algorithm to screen target core markers associated with sepsis. Since the weighted gene co-expression network analysis can cluster genes with similar expression patterns into modules, these modules are associated with sepsis, and through soft threshold optimization in the weighted gene co-expression network analysis, the subsequent screening of the target core markers is more consistent with the scale-free network distribution, reduces noise interference, improves the accuracy of obtaining target core markers associated with sepsis, and subsequently constructs a diagnostic model according to the expression amount of the target core standard material, and diagnoses the patient's serum through the diagnostic model, achieving the purpose of early diagnosis and accurate detection of child sepsis detection.
[0064] Reference Figure 1 As shown in the figure, the application discloses a sepsis data processing method, which mainly includes the following steps:
[0065] S101: Obtain first gene expression data of sepsis and second gene expression data of healthy control samples.
[0066] In S101, the first gene expression data of child sepsis (PS) and the second gene expression data of healthy control (HC) samples are obtained from the public gene expression (Gene Expression Omnibus, GEO) database. The specific data set used is GSE66099, which contains 199 sepsis patients and 47 healthy control group whole blood samples.
[0067] Among them, the number of the above 199 sepsis patients and 47 healthy control group whole blood samples is only an example, and the application does not make specific limitations.
[0068] S102: Data preprocessing and batch effect correction are performed on the first gene expression data and the second gene expression data.
[0069] The first gene expression data and the second gene expression data are preprocessed and batch effect corrected (SVA package) to reduce the influence of technical bias on the analysis results.
[0070] Batch effect correction can be performed by surrogate variable analysis (SVA) package.
[0071] The process of data preprocessing of the first gene expression data and the second gene expression data is as shown in A1-A5.
[0072] A1: calling the ComBat function of the sva package to calculate the mean expression level and the mean difference between batches in the first gene expression data and the second gene expression data.
[0073] wherein the mean difference between batches can be calculated by batch effect estimation.
[0074] A2: adjusting the mean difference between batches by empirical Bayes method.
[0075] A3: calculating the adjustment parameter of the adjusted mean difference between batches.
[0076] The adjustment parameter of the adjusted mean difference between batches is represented by (a, b).
[0077] A4: variance shrinkage is performed on the first gene expression data and the second gene expression data.
[0078] A5: adjusting the first gene expression data after variance shrinkage and the second gene expression data after variance shrinkage by the adjustment parameter to obtain an expression value matrix, so as to complete the process of data preprocessing of the first gene expression data and the second gene expression data.
[0079] The process of batch effect correction of the first gene expression data and the second gene expression data is shown as B1-B2.
[0080] B1: function correction is performed on the first gene expression data and the second gene expression data.
[0081] In B1, the first gene expression data, the second gene expression data and the batches are functionally corrected by a technical bias elimination mechanism.
[0082] B2: eliminating the batch variation source of the first gene expression data after function correction and the batch variation source of the second gene expression data after function correction, so as to complete the process of batch effect correction of the first gene expression data and the second gene expression data.
[0083] While retaining the biologically relevant variation, the technical batch difference is eliminated.
[0084] S103: a plurality of (such as 1202) differential expression genes (DEGs) are screened out from the processing corrected first gene expression data and the processing corrected second gene expression data by differential expression analysis method.
[0085] Wherein, the plurality of differential expression genes at least include a plurality of up-regulated genes and a plurality of down-regulated genes.
[0086] In S103, differential expression analysis is performed on the first gene expression data after processing correction and the second gene expression data after processing correction by using the R language package (Linear Models for Microarray Data, limma) under the Bioconductor project, and a plurality of differentially expressed genes are screened out, such as 1202 genes.
[0087] Specifically, the plurality of differentially expressed genes (DEGs) (such as 1202 genes) are screened out from the first gene expression data after processing correction and the second gene expression data after processing correction, as shown in C1-C5.
[0088] C1: In the data preparation process, a design matrix representing sample grouping is constructed; wherein the design matrix representing sample grouping at least includes a sepsis sample grouping and a healthy control sample grouping.
[0089] C2: A contrast matrix is set; wherein the contrast matrix describes the contrast relationship between each gene expression data.
[0090] In C2, the comparison relationship between the sepsis sample grouping and the healthy control sample grouping is specified by setting the contrast matrix.
[0091] C3: Linear model fitting is performed on the first gene expression data after processing correction and the second gene expression data after processing correction to obtain a statistical quantity of gene expression difference and a probability value (P value); wherein the probability value is the probability of occurrence of a sample result or an extreme result under a preset condition.
[0092] The preset condition is that the original hypothesis is true.
[0093] The probability value is represented by P value. Under the condition that the original hypothesis is true, the probability of observing a sample result or a more extreme result.
[0094] C4: A plurality of up-regulated genes are screened out from the first gene expression data after processing correction by using the sepsis sample grouping, the contrast matrix, the statistical quantity of gene expression difference, and the probability value (P value).
[0095] C5: A plurality of down-regulated genes are screened out from the second gene expression data after processing correction by using the healthy control sample grouping, the contrast matrix, the statistical quantity of gene expression difference, and the probability value (P value).
[0096] In the screening process, the screening condition of P value less than 0.05 and |log2FC|≥1 is set as the screening standard of DEGs. Finally, 1202 DEGs are screened out, including 612 up-regulated genes and 590 down-regulated genes. These DEGs can be displayed by a heat map and a volcano plot, as shown in Figure 2The heat map is shown to show the difference in gene expression between the sepsis group and the healthy group, providing basic data for subsequent analysis.
[0097] Figure 2 The DEGs analysis and enrichment analysis of GSE131761 dataset are involved. Figure 2 A in FIG. 1 indicates a heat map; Figure 2 B in FIG. 1 indicates a volcano plot; Figure 2 C in FIG. 1 indicates KEGG signal pathway enrichment analysis; Figure 2 D in FIG. 1 is the biological process (BP) in the enrichment analysis (Gene Ontology, GO); Figure 2 E in FIG. 1 is the cellular component (CC) in the GO analysis; F is the molecular function (MF) in the GO analysis.
[0098] S104: Screening multiple key module genes related to sepsis and initial hub genes related to sepsis by WGCNA and DEGs.
[0099] In S104, the gene co-expression network is further constructed by WGCNA and DEGs to screen the key modules and initial hub genes related to sepsis. The initial hub genes are a larger range of initial hub genes that are not processed by the preset machine learning algorithm combination.
[0100] The process of screening multiple key module genes related to sepsis and initial hub genes related to sepsis is shown in D1-D5.
[0101] D1: Set a soft threshold (power) by a soft threshold function.
[0102] In D1, a soft threshold function is used to select an appropriate soft threshold. The soft threshold is used to determine the correlation between genes and ensure the size freedom of the network, so as to establish a gene co-expression network that meets the size freedom.
[0103] The pickSoftThreshold() function is executed, the network topology characteristics at different power values are calculated by the pickSoftThreshold() function, the minimum power value that makes the network approximate scale-free (R2>0.8) is selected, and the minimum power value that makes the network approximate scale-free (R2>0.8) is determined as the optimal power value and the minimum power value is output to construct the network. The scale-free gene co-expression network is constructed.
[0104] D2: A plurality of gene modules are identified from the DEGs by hierarchical clustering analysis and dynamic tree cutting method; wherein the gene module is a group of genes whose expression profiles show consistent changes under given conditions.
[0105] The plurality of gene modules represent the high similarity of genes in expression.
[0106] The hierarchical clustering analysis code is as follows:
[0107] geneTree=hclust(as.dist(1-TOM), method="average").
[0108] The execution code of the dynamic tree cutting method is as follows:
[0109] dynamicMods=cutreeDynamic(dendro=geneTree, method="hybrid").
[0110] D3: Calculate the first correlation of the characteristic genes in the plurality of gene modules with the first gene expression data after processing correction, and calculate the second correlation of the characteristic genes in the plurality of gene modules with the second gene expression data after processing correction.
[0111] In D3, the first correlation of the characteristic genes in the plurality of gene modules with PS is calculated, and the second correlation of the characteristic genes in the plurality of gene modules with HC is calculated.
[0112] D4: According to the first correlation, the second correlation and the soft threshold, a plurality of key module genes related to sepsis are determined.
[0113] Among them, the plurality of key module genes related to sepsis are module characteristic genes with correlation greater than the soft threshold, i.e. dark green modules. The dark green modules are significantly related to the clinical characteristics of sepsis.
[0114] The specific process of determining a plurality of gene modules related to sepsis can be referred to as shown in Figure 3 .
[0115] Figure 3 In some embodiments, the core genes (Hub genes) are screened by WGCNA, i.e. the multiple key module genes described above.
[0116] Figure 3 A in the above figure represents connectivity test; B represents independence test; C represents gene clustering; D represents matrix heat map of module correlation expression; E represents gene clustering grouping; F-G represents module analysis, F represents correlation analysis of two groups and 28 module genes, and G describes the scatter plot of the relationship between MM and GS in the olive green module.
[0117] It should be noted that WGCNA (Weighted Gene Co-expression Network Analysis) is a method of constructing co-expression network based on gene expression data, which analyzes the co-expression relationship between genes (i.e. the similarity of gene expression patterns in different samples) to mine potential biological information. WGCNA method reveals how genes work together in biological processes by dividing genes into different modules, and ultimately identifies gene modules related to specific diseases, traits or clinical characteristics.
[0118] Hub Genes:
[0119] In the gene co-expression network, Hub genes refer to genes with very high connectivity in the co-expression network. They are usually located in the core position of the network and have more connections (i.e. co-expression relationship with other genes). Due to their key role in the regulatory network, Hub genes are usually genes with important biological functions. The screening of Hub genes is a core task in WGCNA. Its goal is to identify genes with important roles in biological functions, disease prediction, etc. from the constructed gene co-expression network.
[0120] D5: From the multiple key module genes related to sepsis, screen out genes meeting the conditions as the preset number of initial core markers related to sepsis.
[0121] The preset number includes but is not limited to 215.
[0122] In D5, for each key module gene, the connectivity of each key module gene is calculated, and the gene with the highest connectivity in the module is selected as the hub gene. The connectivity of the gene meeting the preset condition is selected from the connectivity of each gene as the initial core marker related to sepsis.
[0123] The preset condition is that the average connectivity is above 100.
[0124] S105: Perform intersection analysis on the plurality of DEGs and the plurality of key module genes to obtain a plurality of intersection genes.
[0125] For example, 180 intersection genes are obtained by performing intersection analysis on the 215 module genes screened by WGCNA and the 1202 DEGs screened.
[0126] S106: Screen the target core marker from the initial core marker by the preset machine learning algorithm combination and the plurality of intersection genes.
[0127] Among them, the target core marker includes 4 hub genes, namely TLR5, S100A9, HSDL2 and SLC2A3. These genes show high performance in the diagnosis of sepsis.
[0128] In S106, the target core marker is screened from the initial core marker by the preset machine learning algorithm combination composed of LASSO regression algorithm, SVM-RFE algorithm and random forest (RF) algorithm and the plurality of intersection genes.
[0129] In order to further confirm which genes are most related to the diagnosis of sepsis, the present application uses three machine learning algorithms: LASSO regression algorithm, SVM-RFE algorithm (support vector machine recursive feature elimination) and random forest (RF) algorithm to form a preset machine learning algorithm combination. The target core marker is screened from the initial core marker by the preset machine learning algorithm combination and the plurality of intersection genes.
[0130] LASSO regression algorithm analysis:
[0131] LASSO regression is used for variable selection, and finally 19 candidate genes are screened. These genes have strong correlation in the LASSO model and can be used as potential diagnostic markers for sepsis.
[0132] a. Data preparation: Prepare gene expression data and the first target variable (sepsis diagnosis result).
[0133] b. Data standardization: Standardize the first gene expression data of sepsis.
[0134] c. Train LASSO model:
[0135] d. Screen candidate genes: Extract the regression coefficient and screen the genes with non-zero coefficient.
[0136] e. Output result: Output the list of screened candidate genes.
[0137] SVM-RFE algorithm:
[0138] Using the Support Vector Machine Recursive Feature Elimination algorithm, 12 genes were further selected, which also showed diagnostic prediction ability for sepsis in the SVM model.
[0139] a. Prepare data: Input includes gene expression matrix and second target variable (sepsis diagnosis label).
[0140] b. Standardize data: Standardize the data to avoid the influence of different gene scales.
[0141] c. Initialize SVM model: Use linear SVM model as the base model.
[0142] d. Perform Recursive Feature Elimination (SVM-RFE).
[0143] e. Automatically perform recursive feature elimination.
[0144] f. Calculate the influence of each feature, evaluated by the weight of the support vector machine model or the distance of the support vector.
[0145] g. Recursively remove the least important genes until the most important genes are selected.
[0146] h. Output selected features: The final output is the selected genes that are most useful for sepsis diagnosis, which can be used as biomarkers for subsequent analysis or clinical verification.
[0147] Random Forest Algorithm:
[0148] Using the Random Forest algorithm to further analyze the genes selected above. By calculating the relationship between the number of classification trees and the error rate, 20 genes with higher correlation were determined.
[0149] 1. Input data:
[0150] Input data set: In this scheme, the input data is 215 candidate genes related to sepsis (i.e. key module genes) selected by WGCNA and 1202 DEGs selected previously. Through intersection analysis of the two data sets, 180 intersection genes are obtained as the final input feature set.
[0151] 2. Random Forest Algorithm Process:
[0152] Training data preparation: 180 intersection genes are used as input features, and the class label of each sample is sepsis (PS) or healthy control (HC). Each sample contains the expression of genes as features.
[0153] Build a random forest:
[0154] Data Sampling: Randomly extract multiple subsets of data from the original training data.
[0155] Training Multiple Decision Trees: For each subset of data, build an independent decision tree. At each split, randomly select a subset of features instead of using all features, increasing model diversity and preventing overfitting.
[0156] Ensemble Decision Trees: Integrate the prediction results of multiple decision trees to obtain the final classification result.
[0157] Model Training: Random Forest algorithm trains multiple decision trees, evaluates the relationship between genes and sepsis based on the integration results of these trees, and finally selects a set of genes with the highest predictive value.
[0158] 3. Output:
[0159] Selected Genes: Through the trained Random Forest model, 20 candidate genes with strong correlation with sepsis diagnosis are selected.
[0160] Feature Importance: Random Forest can also evaluate the contribution of each feature (gene) to the prediction result, outputting the importance score of each gene.
[0161] Finally, the intersection analysis of these methods filters out four important hub genes, TLR5, S100A9, HSDL2, and SLC2A3, which are the target core markers. These target core markers show high efficiency in the diagnosis of sepsis. The specific preset machine algorithm combination is shown in Figure 4 .
[0162] Figure 4 Among them, machine learning algorithm A: 180 common genes of DEGs and WGCNA are obtained by Venn diagram analysis; Figure 4 B in the PPI-protein interaction analysis; Figure 4 C-E in the lasso, random vector machine (SVM-RFE), and random forest (RF) machine learning algorithms; Figure 4 F in the Venn diagram to obtain four hub genes (TLR5, S100A9, HSDL2, SLC2A3); Figure 4 G in the ROC curve verification of the four hub genes in four external data; Figure 4 H in the gene expression of the four hub genes in the experimental group and the control group in the four validation sets.
[0163] The target core markers were validated by using multiple external datasets (such as GSE13904, GSE145227, GSE26378, and GSE26440). ROC curve analysis showed that the AUC values of the four hub genes were all over 0.75, indicating that these genes have high diagnostic efficacy and predictive accuracy.
[0164] 1. Input data:
[0165] The input data of the external datasets contain gene expression data of sepsis patients and healthy controls. Each dataset contains gene expression profiles (such as RNA-Seq or microarray data) and each dataset has samples from sepsis patients (PS) and healthy controls (HC).
[0166] Specific dataset:
[0167] GSE13904: A gene expression dataset containing sepsis patients and healthy controls.
[0168] GSE145227: Another external dataset containing patients with sepsis and healthy controls.
[0169] GSE26378: a sepsis-related dataset containing samples with different clinical characteristics.
[0170] GSE26440: Another public dataset related to sepsis used to validate diagnostic models.
[0171] 2. Data preprocessing:
[0172] Before using these external datasets, the following preprocessing steps are performed:
[0173] Data normalization: Because batch effects or experimental platform differences may exist between different datasets, gene expression data needs to be normalized to ensure data consistency. In addition, batch effect correction is performed on the data to ensure the reliability of the results.
[0174] Gene screening and matching: Because gene expression may vary between datasets, ensure that the genes used in these external datasets match the four identified hub genes (TLR5, S100A9, HSDL2, and SLC2A3). If certain genes are missing from the dataset, a gene mapping strategy can be used to match the relevant genes.
[0175] 3. ROC curve analysis:
[0176] The ROC (Receiver Operating Characteristic) curve is used to evaluate the diagnostic ability of different gene combinations for sepsis. The specific steps are as follows:
[0177] Gene expression extraction: Extract the expression levels of the 4 hub genes (TLR5, S100A9, HSDL2, SLC2A3) from the external datasets.
[0178] Build classification model: Based on the expression levels of the 4 hub genes, build a diagnostic model. A simple weighted sum model (e.g., weighted sum of all gene expression values, or using a logistic regression model, etc.) can be used.
[0179] Calculate AUC value: Using the sample labels of the sepsis group (PS) and healthy control group (HC) in the external datasets, evaluate the diagnostic performance of the model by calculating the area under the ROC curve (AUC).
[0180] The closer the AUC value is to 1, the higher the diagnostic performance of the model.
[0181] Evaluate the sensitivity, specificity, and accuracy of the model.
[0182] Draw ROC curve: For each dataset, draw the ROC curve and calculate the AUC value.
[0183] The ROC curve can be drawn and the AUC value can be calculated using the pROC package in R.
[0184] 4、Output results:
[0185] AUC value: For each external dataset, calculate the AUC value of the 4 hub gene combination. AUC value is commonly used to measure the performance of the model, and AUC value greater than 0.75 is usually considered as a good diagnostic performance indicator.
[0186] GSE13904: AUC value of 0.76, showing good predictive performance.
[0187] GSE145227: AUC value of 0.80, further verifying the diagnostic performance of the model.
[0188] GSE26378: AUC value of 0.78, indicating that the model performs stably in different samples.
[0189] GSE26440: AUC value of 0.82, further supporting the diagnostic value of the 4 hub genes in sepsis.
[0190] Statistical analysis: By comparing the AUC values of different datasets, check the stability and generalization ability of the model in different external datasets.
[0191] The protein expression levels of the four hub genes (TLR5, S100A9, HSDL2, SLC2A3) in the serum of sepsis patients were verified by ELISA experiments. The ELISA results show that the four genes have significantly high expression in sepsis patients (46 cases), and the difference is statistically significant compared with the healthy control group (42 cases) (P<0.05). The specific ELISA experiment verification is shown in Figure 5 .
[0192] Figure 5 The clinical sample elisa detects the protein expression levels of the four hub genes (TLR5, S100A9, HSDL2, SLC2A3).
[0193] A diagnostic model is constructed according to the expression amount of the target core standard, and the serum of the patient is diagnosed through the diagnostic model.
[0194] The key points of the present application are as follows:
[0195] 1. Hub gene screening method combining multiple algorithms:
[0196] The present application proposes a children's sepsis biomarker screening and diagnosis method combining WGCNA method and machine learning algorithm, which successfully screens hub genes with high diagnostic efficiency through multiple levels of data analysis and verification steps, and constructs an effective sepsis diagnosis model.
[0197] The present application successfully screens four key hub genes (TLR5, S100A9, HSDL2, SLC2A3) related to children's sepsis by combining WGCNA and machine learning algorithm, which can improve the screening accuracy and overcome the limitations of single algorithm.
[0198] 2. Construction and verification of high-efficiency diagnosis model:
[0199] Based on the screened target core marker, the present application constructs a high-efficiency diagnosis model, and proves that the diagnosis model has good diagnostic efficiency through multi-data set verification. The diagnosis model provides a new biomarker combination for early diagnosis of sepsis, and has a wide clinical application prospect.
[0200] 3. Correlation research between immune cells and hub genes:
[0201] The present application deeply explores the relationship between hub genes and immune cell infiltration, and reveals that these genes may play a role in the occurrence and development of sepsis by regulating immune response, which provides a new idea for the immunotherapy of sepsis.
[0202] 4. Clinical verification:
[0203] The significant differences of TLR5, S100A9, HSDL2 and SLC2A3 in the serum of sepsis patients are verified by ELISA experiments, which provides experimental evidence for the application of the biomarkers in clinic.
[0204] The method embodiment of the present application has the following beneficial effects: the target core markers associated with sepsis are screened out by combining the weighted gene co-expression network analysis method with the preset machine learning algorithm, the genes with similar expression patterns are clustered into modules by the weighted gene co-expression network analysis, these modules are associated with sepsis, and the subsequent screening of the target core markers is more in line with the scale-free network distribution, the noise interference is reduced, the accuracy of obtaining the target core markers associated with sepsis is improved, and the subsequent diagnosis model can be constructed according to the expression amount of the target core markers, and the serum of patients is diagnosed through the diagnosis model, so as to realize the purposes of early diagnosis and accurate detection of child sepsis detection.
[0205] Based on the above embodiments Figure 1 The disclosed processing method of sepsis data, the embodiment of the present application also discloses a processing device of sepsis data, as shown in the figure, the processing device of sepsis data comprises: Figure 6 As shown in the figure, the processing device of sepsis data comprises:
[0206] The acquisition unit 601 is configured to acquire first gene expression data of sepsis and second gene expression data of healthy control samples.
[0207] The processing correction unit 602 is configured to perform data preprocessing and batch effect correction on the first gene expression data and the second gene expression data.
[0208] The first screening unit 603 is configured to screen a plurality of differentially expressed genes from the first gene expression data and the second gene expression data after processing correction by a differential expression analysis method.
[0209] The second screening unit 604 is configured to screen a plurality of key module genes related to sepsis and an initial core marker related to sepsis from the differentially expressed genes by a weighted gene co-expression network analysis method.
[0210] The intersection analysis unit 605 is configured to perform intersection analysis on the plurality of differentially expressed genes and the plurality of key module genes to obtain a plurality of intersection genes.
[0211] The third screening unit 606 is configured to screen a target core marker from the initial core marker by combining a plurality of intersection genes with a preset machine learning algorithm.
[0212] Furthermore, the processing and correction unit 602 that performs data preprocessing on the first gene expression data and the second gene expression data includes:
[0213] The first calculation module is used to call the ComBat function of the sva package to calculate the mean expression level of the first gene expression data and the second gene expression data in all samples and the mean difference between batches;
[0214] The first adjustment module is used to adjust the mean difference between batches by using the empirical Bayesian leaf method;
[0215] A second calculation module is used to calculate an adjustment parameter of the adjusted mean difference between batches;
[0216] a variance shrinkage module, configured to perform variance shrinkage on the first gene expression data and the second gene expression data;
[0217] The first adjustment module is used to adjust the first gene expression data after variance shrinkage and the second gene expression data after variance shrinkage by adjusting parameters to obtain an expression value matrix to complete the process of data preprocessing for the first gene expression data and the second gene expression data.
[0218] Furthermore, the processing correction unit 602 for performing batch effect correction on the first gene expression data and the second gene expression data includes:
[0219] a function correction module, configured to perform function correction on the first gene expression data and the second gene expression data;
[0220] The elimination module is used to eliminate the batch variation source of the first gene expression data after function correction and the batch variation source of the second gene expression data after function correction, so as to complete the process of batch effect correction of the first gene expression data and the second gene expression data.
[0221] Furthermore, the first screening unit 603 includes:
[0222] A construction module is used to construct a design matrix to represent sample grouping; wherein the design matrix represents sample grouping including at least sepsis sample grouping and healthy control sample grouping;
[0223] The first setting module is used to set a comparison matrix; wherein the comparison matrix describes the comparison relationship between each gene expression data;
[0224] A fitting module is used to perform linear model fitting on the processed and corrected first gene expression data and the processed and corrected second gene expression data to obtain statistics and probability values of gene expression differences; wherein the probability value is the probability of a sample result or an extreme result occurring under a preset condition;
[0225] The first screening module is configured to screen a plurality of up-regulated genes from the first gene expression data after processing correction by a sepsis sample grouping, a comparison matrix, a statistical quantity of gene expression difference, and a probability value.
[0226] The second screening module is configured to screen a plurality of down-regulated genes from the second gene expression data after processing correction by a healthy control sample grouping, a comparison matrix, a statistical quantity of gene expression difference, and a probability value.
[0227] Further, the second screening unit 602 comprises:
[0228] The second setting module is configured to set a soft threshold value by a soft threshold function.
[0229] The identification module is configured to identify a plurality of gene modules from the differentially expressed genes by a hierarchical clustering analysis and a dynamic tree cutting method; wherein the gene module is a group of genes whose expression profiles show consistent changes under a given condition.
[0230] The third calculation module is configured to calculate a first correlation of a feature gene in the plurality of gene modules with the first gene expression data after processing correction, and a second correlation of the feature gene in the plurality of gene modules with the second gene expression data after processing correction.
[0231] The determination module is configured to determine a plurality of key module genes related to sepsis according to the first correlation, the second correlation, and the soft threshold value.
[0232] The third screening module is configured to screen a gene meeting a condition from the plurality of key module genes related to sepsis as a preset number of initial core markers related to sepsis.
[0233] Further, the third screening unit 604 is specifically configured to screen a target core marker from the initial core markers by a preset machine learning algorithm combination composed of a LASSO regression algorithm, a SVM-RFE algorithm, and a random forest algorithm, and a plurality of intersection genes.
[0234] Further, the sepsis data processing apparatus further comprises:
[0235] The construction unit is configured to construct a diagnosis model according to the expression amount of the target core marker, and to diagnose the patient's serum by the diagnosis model.
[0236] The device embodiment of the present application has the following beneficial effects: the target core marker associated with sepsis is screened out by combining the weighted gene co-expression network analysis method with the preset machine learning algorithm. Since the weighted gene co-expression network analysis can cluster genes with similar expression patterns into modules, these modules are associated with sepsis, and the subsequent screening of the target core marker is more in line with the scale-free network distribution, reduces noise interference, and improves the accuracy of obtaining the target core marker associated with sepsis. Subsequently, a diagnostic model can be constructed according to the expression amount of the target core standard material, and the patient's serum is diagnosed through the diagnostic model, thereby achieving the purpose of early diagnosis and accurate detection of child sepsis detection.
[0237] The present application also provides a storage medium, which comprises stored instructions, wherein the instructions control the device where the storage medium is located to perform the processing method of sepsis data as described above when the instructions are executed.
[0238] The present application also provides an electronic device, the structural schematic diagram of which is shown in Figure 7 The electronic device specifically comprises a memory 701 and one or more instructions 702, wherein the one or more instructions 702 are stored in the memory 701 and are configured to be executed by one or more processors 703 to execute the processing method of sepsis data as described above.
[0239] For the foregoing method embodiments, in order to simply describe, they are all expressed as a combination of a series of actions, but those skilled in the art should know that the present application is not limited to the action sequence described, because according to the present application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should know that the embodiments described in the specification all belong to preferred embodiments, and the actions and modules involved are not necessarily necessary for the present application.
[0240] It should be noted that each embodiment in the present specification is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same and similar parts between the embodiments can be referred to. For device embodiments, since they are basically similar to method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the method embodiment.
[0241] The steps in the method of each embodiment of the present application can be adjusted in sequence, combined and deleted according to actual needs.
[0242] Finally, it should be noted that the terminology used herein, such as first and second, is merely used for the convenience of the reader and is not intended to, and should not be interpreted to, imply or suggest any actual relationship between, or order of, such entities or actions.
[0243] The above description of disclosed embodiments is merely intended to provide an overview of the application. Various modifications as well as the numerous details of implementation are already apparent to those skilled in the art and can be made without departing from the scope of the application. Therefore, the application is not to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
[0244] The above description is merely intended to provide an overview of the application. Various modifications and improvements will be readily apparent to those of ordinary skill in the art, and the application is not to be limited to the specific embodiments described herein. Rather, the scope of the application is to be determined by the claims which are to be accorded their broadest reasonable interpretation, which can incorporate any and every novel feature and combination of features.
Claims
1. A method for processing sepsis data, characterized in that: include: obtaining first gene expression data of sepsis samples and second gene expression data of healthy control samples; performing data preprocessing and batch effect correction on the first gene expression data and the second gene expression data; Screening a plurality of differentially expressed genes from the processed and corrected first gene expression data and the processed and corrected second gene expression data by differential expression analysis; Through weighted gene co-expression network analysis and the differentially expressed genes, multiple key module genes related to sepsis and initial core markers related to sepsis were screened out; Performing intersection analysis on the multiple differentially expressed genes and the multiple key module genes to obtain multiple intersection genes; Target core markers are screened out from the initial core markers by combining a preset machine learning algorithm and the multiple intersection genes.
2. The method according to claim 1, characterized in that Performing data preprocessing on the first gene expression data and the second gene expression data includes: Call the ComBat function of the sva package to calculate the mean expression levels of the first gene expression data and the second gene expression data in all samples and the mean differences between batches; The mean differences between batches were adjusted by the empirical Bayesian method; Calculate the adjustment parameter for the adjusted mean differences between batches; performing variance shrinkage on the first gene expression data and the second gene expression data; The first gene expression data after variance shrinkage and the second gene expression data after variance shrinkage are adjusted by the adjustment parameters to obtain an expression value matrix, thereby completing the data preprocessing process for the first gene expression data and the second gene expression data.
3. The method according to claim 1, characterized in that Performing batch effect correction on the first gene expression data and the second gene expression data, comprising: performing function correction on the first gene expression data and the second gene expression data; Eliminate the batch variation source of the function-corrected first gene expression data and the batch variation source of the function-corrected second gene expression data to complete the process of performing batch effect correction on the first gene expression data and the second gene expression data.
4. The method according to claim 1, wherein The plurality of differentially expressed genes include at least a plurality of up-regulated genes and a plurality of down-regulated genes, and the plurality of differentially expressed genes are screened out from the processed and corrected first gene expression data and the processed and corrected second gene expression data by the differential expression analysis method, comprising: Constructing a design matrix to represent sample grouping; wherein the design matrix to represent sample grouping includes at least a sepsis sample grouping and a healthy control sample grouping; Setting a comparison matrix; wherein the comparison matrix describes the comparison relationship between each gene expression data; Performing linear model fitting on the processed and corrected first gene expression data and the processed and corrected second gene expression data to obtain a statistic and a probability value of the gene expression difference; wherein the probability value is the probability of the sample result or extreme result occurring under a preset condition; Screening the multiple up-regulated genes from the processed and corrected first gene expression data by using the sepsis sample grouping, the comparison matrix, the statistic of the gene expression difference, and the probability value; The plurality of down-regulated genes are screened out from the processed and corrected second gene expression data by using the healthy control sample grouping, the comparison matrix, the statistics of the gene expression differences and the probability value.
5. The method according to claim 1, wherein The weighted gene co-expression network analysis method and the differentially expressed genes are used to screen out multiple key module genes related to sepsis and initial core markers related to sepsis, including: Set the soft threshold through the soft threshold function; Identifying multiple gene modules from the differentially expressed genes through hierarchical clustering analysis and dynamic tree cutting methods; wherein the gene module is a group of genes whose expression profiles show consistent changes under given conditions; Calculating a first correlation between the characteristic genes in the plurality of gene modules and the processed and corrected first gene expression data, and calculating a second correlation between the characteristic genes in the plurality of gene modules and the processed and corrected second gene expression data; determining a plurality of key module genes associated with sepsis according to the first correlation, the second correlation, and the soft threshold; Qualified genes were screened from multiple key module genes related to sepsis as a preset number of initial core markers related to sepsis.
6. The method according to claim 1, characterized in that The method of screening target core markers from the initial core markers by combining a preset machine learning algorithm and the plurality of intersection genes comprises: The target core markers are screened out from the initial core markers by using a preset machine learning algorithm combination consisting of a LASSO regression algorithm, an SVM-RFE algorithm, and a random forest algorithm and the multiple intersection genes.
7. The method according to claim 1, characterized in that Also includes: A diagnostic model is constructed based on the expression level of the target core standard, and the patient serum is diagnosed using the diagnostic model.
8. A device for processing sepsis data, characterized in that: include: an acquiring unit, configured to acquire first gene expression data of sepsis and second gene expression data of a healthy control sample; a processing and correction unit, configured to perform data preprocessing and batch effect correction on the first gene expression data and the second gene expression data; A first screening unit is configured to screen out a plurality of differentially expressed genes from the processed and corrected first gene expression data and the processed and corrected second gene expression data by using a differential expression analysis method; A second screening unit is used to screen out multiple key module genes related to sepsis and initial core markers related to sepsis through weighted gene co-expression network analysis and the differentially expressed genes; An intersection analysis unit, configured to perform intersection analysis on the plurality of differentially expressed genes and the plurality of key module genes to obtain a plurality of intersection genes; The third screening unit is used to screen out target core markers from the initial core markers by combining a preset machine learning algorithm and the multiple intersection genes.
9. A storage medium, characterized in that: The storage medium includes stored instructions, wherein when the instructions are executed, the device where the storage medium is located is controlled to execute the method for processing sepsis data according to any one of claims 1 to 7.
10. An electronic device, characterized in that: The system comprises a memory and one or more instructions, wherein the one or more instructions are stored in the memory and configured to be executed by one or more processors to perform the method for processing sepsis data according to any one of claims 1 to 7.