Benzene-caused blood abnormality marker, evaluation model and construction method and application of benzene-caused blood abnormality marker and evaluation model

Abstract

The invention relates to the technical field of biology, in particular to a benzene-induced blood abnormality marker, an evaluation model and a construction method and application of the benzene-induced blood abnormality marker and the evaluation model, and provides the following scheme that the method comprises the steps that S1, principal component analysis is conducted on benzene damage expression profile data sets, then preliminary screening conditions are set, and differential expression genes between the benzene damage expression profile data sets are screened out; s2, carrying out GO, KEGG and GSEA enrichment analysis on the differentially expressed gene to obtain an analysis result; s3, different machine learning algorithms are used for screening characteristic genes from analysis results, intersection is taken, key genes are obtained, then verification analysis is conducted on the key genes, and benzene damage marker genes are obtained; and S4, training and testing a plurality of learning models by using the benzene injury marker gene, and selecting a proper learning model as a benzene-induced blood abnormality evaluation model according to a test result of the learning model. According to the invention, a theoretical basis can be provided for early intervention of benzene toxicity, and a more comprehensive and sensitive model can be provided for risk assessment.

Description

Technical Field

[0001] The present invention relates to the field of biotechnology, and in particular to a benzene-induced blood abnormality marker, an evaluation model, a construction method and an application thereof. Background Art

[0002] Benzene, as a common organic hydrocarbon industrial solvent, is widely used in the pharmaceutical, pesticide, medicine, shoemaking and other industries, causing widespread occupational exposure. Because it can cause various blood-related diseases, it is listed as a Class I carcinogen by the International Agency for Research on Cancer (IARC1982). Occupational benzene poisoning is divided into two types: acute and chronic poisoning. Among them, the most common is occupational chronic benzene poisoning. High-concentration benzene exposure can cause serious blood system diseases, such as pancytopenia, aplastic anemia and leukemia; in addition, studies have shown that long-term low-concentration benzene exposure can also induce blood toxicity. Benzene has no safe exposure limit, and even exposure to lower concentrations may cause blood malignancies. Therefore, it is particularly important to actively explore biomarkers of low-concentration benzene exposure, providing a reference for screening and early intervention treatment of high-risk populations exposed to benzene; and machine learning is a computer program that can learn from experience in certain tasks and performance metrics. It is widely used in bioinformatics due to its adaptability and self-regulation characteristics, and is used by researchers to explore the potential mechanisms, potential biomarkers and treatment targets of various diseases.

[0003] To this end, the present invention proposes a benzene-induced blood abnormality marker, an evaluation model, a construction method and an application thereof. Summary of the invention

[0004] In order to effectively conduct risk assessment of benzene poisoning and identify early biomarkers of abnormal benzene poisoning, the present invention proposes a benzene-induced blood abnormality marker, an evaluation model, a construction method and application thereof.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] The first aspect of the present invention provides a method for constructing a benzene-induced blood abnormality assessment model, comprising the following steps:

[0007] S1: First, principal component analysis was performed on the benzene damage expression profile dataset, and then the initial screening conditions were set to screen out the differentially expressed genes between the benzene damage expression profile datasets;

[0008] Preferably, the benzene damage expression profile dataset is selected from the benzene damage expression profile datasets GSE9569 and GSE21862 in the GEO database (Gene Expression Omnibus);

[0009] S2: Perform GO, KEGG, and GSEA enrichment analysis on the differentially expressed genes to obtain the analysis results;

[0010] S3: Use different machine learning algorithms to screen out characteristic genes from the analysis results and take the intersection to obtain key genes, then perform verification analysis on the key genes to obtain benzene damage marker genes;

[0011] S4: Use benzene damage marker genes to train and test multiple learning models, and select a suitable learning model as a benzene-induced blood abnormality assessment model based on the test results of the learning model.

[0012] In some embodiments, setting the initial screening conditions in S1 includes: using the R software "limma" package, with the significance value P<0.05 and |logfoldchange|≥0.5 as the initial screening conditions; after screening out the differentially expressed genes, drawing a heat map with the "pheatmap" package and drawing a volcano map with the "ggplot2" package.

[0013] In some embodiments, the step S2 includes using the "ClusterProfiler" package in R software to perform GO, KEGG, and GSEA enrichment analysis on the differentially expressed genes, with a significance value of p < 0.05 as the standard, and using a bubble chart to display the enrichment analysis results.

[0014] In some embodiments, the machine learning algorithms in S3 include: lasso regression algorithm, SVM-RFE support vector machine recursive feature elimination algorithm and random forest algorithm.

[0015] In some embodiments, the step of S3 includes:

[0016] S3.1: The characteristic genes were screened using three machine learning algorithms: lasso regression algorithm, SVM-RFE support vector machine recursive feature elimination algorithm and random forest algorithm through R software. Then the venn package was used to take the intersection of the characteristic genes screened by the three machine learning algorithms to obtain the key genes.

[0017] S3.2: Use R software to draw the receiver operating curve of the obtained key genes, calculate the area under the curve AUC value, the larger the AUC value, the better the diagnostic performance; and draw a violin plot for observation and analysis;

[0018] S3.3: Select research subjects and divide them into benzene-exposed and benzene-non-exposed groups. Perform population verification and statistical analysis on the correspondence between key genes and benzene poisoning to obtain benzene damage marker genes.

[0019] Preferably, SPSS27.0 software is used for data analysis. For data that conform to the normal distribution, the data are expressed, and the differences between the groups are compared using the independent sample t-test analysis; for data that do not conform to the normal distribution, the median and quartiles are used to express, and the differences between the groups are compared using non-parametric tests, and p < 0.05 is considered to be significantly different.

[0020] Preferably, the benzene damage marker genes include NFKB1 and PTX3.

[0021] In some embodiments, the multiple learning models in S4 include: a support vector machine model, a Bp neural network model, a Bayesian model and a C5.0 decision tree; the method of selecting a suitable learning model according to the test results of the learning model includes: calculating the accuracy of the test results of each learning model and drawing an ROC curve, and comparing them, and using GraphPad Prism 8.3 software to draw the comparison results to obtain a suitable learning model; preferably, the trained C5.0 decision tree is used as a benzene-induced blood abnormality assessment model.

[0022] The second aspect of the present invention provides a benzene-induced blood abnormality assessment model obtained by the above-mentioned construction method.

[0023] The third aspect of the present invention proposes the application of the above-mentioned benzene-induced blood abnormality assessment model in a blood abnormality detection model in a benzene exposure environment.

[0024] The fourth aspect of the present invention proposes a benzene-induced blood abnormality marker obtained by screening using the above model, characterized in that it includes NFKB1 and PTX3.

[0025] The present invention is based on a benzene-exposed population (benzene-exposed group) and a non-benzene-exposed population (control group). Through detection, it is found that NFKB1 and PTX3 in the benzene-exposed group are higher than those in the control group. Subsequently, according to the "Diagnostic Standards for Occupational Benzene Poisoning" GBZ68-2022, the benzene-exposed group was divided into a blood abnormality group and a blood normal group (control group). It was found that NFKB1, PHACTR1 and PTX3 in the blood abnormality group were higher than those in the control group. Combining the two methods, NFKB1 and PTX3 can be considered as key risk genes for benzene exposure damage. At the same time, the present invention also found that the indirect oxidative damage indicator malondialdehyde (MDA), the DNA damage marker 8-hydroxydeoxyguanosine (8-OhdG) and the benzene internal exposure indicator phenylmercaptouric acid (S-PMA) were significantly changed.

[0026] The fifth aspect of the present invention proposes the use of primers for detecting NFKB1 and / or PTX3 in the preparation of reagents for detecting benzene-induced blood abnormalities.

[0027] Beneficial effects of the present invention:

[0028] 1. The present invention first uses the benzene damage expression spectrum data sets GSE9569 and GSE21862 in the GEO database (gene expression comprehensive database) and uses a learning algorithm to screen and obtain key genes, then establishes a research object, analyzes and verifies the key genes, obtains benzene damage marker genes, uses the benzene damage marker genes to construct training sets and test sets to train and test multiple machine learning models, selects the optimal model as the benzene-induced blood abnormality assessment model, thereby combining bioinformatics with machine learning, and verifies the mechanism. The present invention can provide a theoretical basis for early intervention of benzene toxicity, and can provide a more comprehensive and sensitive model for risk assessment.

[0029] 2. The verification results show that the increase of benzene content in the benzene-exposed population causes the increase of inflammation and oxidative stress levels in the body, which may promote the blood toxicity of benzene;

[0030] 3. In previous studies, most of them focused on the relationship between NFKB and benzene exposure, while there were few studies on the relationship between PTX3 and benzene exposure. This study combined machine learning with benzene damage marker genes to construct a model. In addition to NFKB1, PTX3 was also screened out among the key genes of benzene damage, and its screening sensitivity was high;

[0031] 4. This study first combined principal component analysis, gene enrichment analysis and machine learning algorithms to more comprehensively evaluate benzene damage with high reliability; then, through initial screening conditions and machine learning algorithms, differentially expressed genes and key genes were quickly and effectively screened out, greatly improving the efficiency of the analysis; then, the data we used came from a public database, which has strong replicability and versatility, and can be verified and applied among different laboratories and research teams. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The figure is the analysis result diagram of the benzene damage expression spectrum data set GSE9569 and GSE21862 of the present invention, wherein a and b are two-dimensional diagrams of the first two components in the principal component analysis results, a is before adjustment, and b is after adjustment; c is a volcano diagram of the differentially expressed gene results, red dots are up-regulated genes, green dots are down-regulated genes, and gray dots are genes with no difference in expression; d is a cluster heat map of differentially expressed genes;

[0033] Figure 2It is the functional enrichment result diagram of the differentially expressed genes of the present invention, wherein a is the GO enrichment analysis result, which is divided into three parts, from top to bottom, BP biological process, CC cell component and MF molecular function, and the color corresponds to the degree of enrichment; b is the KEGG enrichment analysis result, the horizontal axis represents the number of genes enriched in the pathway, the vertical axis represents the pathway name, and the color corresponds to the degree of enrichment; c is the GSEA diagram of DEGs in the high expression group; d is the GSEA diagram of DEGs in the low expression group;

[0034] Figure 3 Figure 1 is a diagram of the key gene screening results of the present invention, wherein a and b are diagrams of identifying benzene damage-related genes using LASSO and cox regression analysis; c and d are diagrams of SVM-RFE algorithm analysis results; e is a diagram of the random forest algorithm, the horizontal axis is the number of trees, the vertical axis is the cross-validation error, the red line represents the error of the benzene exposure group, the green line represents the error of the control group, and the black line represents the error of all samples; f is a random forest gene importance ranking diagram, which shows the importance ranking of the top 30 genes; g is a Venn diagram combining the three algorithms of LASSO algorithm, SVM-RFE algorithm and random forest algorithm, the red part represents the LASSO algorithm, the green part reaches the standard RF algorithm, the purple part represents the SVM algorithm, the intersection part represents the key genes identified, and the numbers in the figure represent the number of genes;

[0035] Figure 4 : is the ROC curve diagram of the key gene of the present invention, wherein a is the ROC curve diagram of NFKB1; b is the ROC curve diagram of PHACTR1; c is the ROC curve diagram of PTGS2; d is the ROC curve diagram of PTX3;

[0036] Figure 5 It is a violin plot of the relative expression of different genes in the benzene exposure group and the benzene non-exposure group of the present invention, Control represents the benzene non-exposure group, Treat represents the benzene exposure group, wherein a is a violin plot of the expression of NFKB1; b is a violin plot of the expression of PHACTR1; c is a violin plot of the expression of PTGS2; d is a violin plot of the expression of PTX3;

[0037] Figure 6: This is a statistical analysis comparison diagram of the benzene-exposed group and the benzene-non-exposed group of the present invention, and "*" indicates that an independent sample t-test was used, wherein a is a comparison result diagram of 8-OHdG in the benzene-exposed group and the benzene-non-exposed group; b is a comparison result diagram of BMI in the benzene-exposed group and the benzene-non-exposed group; c is a comparison result diagram of MDA in the benzene-exposed group and the benzene-non-exposed group; d is a comparison result diagram of Neutrophil in the benzene-exposed group and the benzene-non-exposed group; e is a comparison result diagram of NF in the benzene-exposed group and the benzene-non-exposed group. f is the comparison result diagram of KB1 in the benzene exposure group and the benzene non-exposed group; g is the comparison result diagram of PTGS2 in the benzene exposure group and the benzene non-exposed group; h is the comparison result diagram of PTX3 in the benzene exposure group and the benzene non-exposed group; i is the comparison result diagram of PLT in the benzene exposure group and the benzene non-exposed group; j is the comparison result diagram of S-PMA in the benzene exposure group and the benzene non-exposed group; k is the comparison result diagram of WBC in the benzene exposure group and the benzene non-exposed group;

[0038] Figure 7 It is a comparison result diagram of the statistical analysis between the abnormal blood group and the normal blood group of the benzene exposure group of the present invention, and "*" indicates that an independent sample t-test is used, wherein a is a comparison result diagram of 8-OHdG between the abnormal blood group and the normal blood group; b is a comparison result diagram of BMI between the abnormal blood group and the normal blood group; c is a comparison result diagram of MDA between the abnormal blood group and the normal blood group; d is a comparison result diagram of Neutrophi between the abnormal blood group and the normal blood group; e is a comparison result diagram of NFKB1 between the abnormal blood group and the normal blood group; f is a comparison result diagram of PHACTR1 between the abnormal blood group and the normal blood group; g is a comparison result diagram of PTGS2 between the abnormal blood group and the normal blood group; h is a comparison result diagram of PTX3 between the abnormal blood group and the normal blood group; i is a comparison result diagram of PLT between the abnormal blood group and the normal blood group; j is a comparison result diagram of S-PMA between the abnormal blood group and the normal blood group; k is a comparison result diagram of WBC between the abnormal blood group and the normal blood group;

[0039] Figure 8The present invention is a comparison diagram of the prediction of the support vector machine model (SVM), the Bp neural network model (BP), the Bayesian model (NaiveBayes) and the C5.0 decision tree (C5.0DT) for the benzene exposure group and the benzene non-exposure group, wherein (ad) is a comparison result diagram of the prediction of the test group and the training group of the benzene exposure group by the four models. (a, c) are respectively the comparison result diagrams of the AUC and accuracy of the four models for the test group, and (b, d) are respectively the comparison result diagrams of the AUC and accuracy of the four models for the training group. (eh) is a comparison result diagram of the prediction of the test group and the training group of the four models for the benzene non-exposure group. (e, g) are respectively the comparison result diagrams of the AUC and accuracy of the four models for the test group, and (f, h) are respectively the comparison result diagrams of the AUC and accuracy of the four models for the training group.

[0040] Fig. 9 The present invention provides ROC curve diagrams of the four machine learning models, namely, support vector machine model (SVM), Bp neural network model (BP), Bayesian model (Naive Bayes) and C5.0 decision tree (C5.0DT) model, and weight diagrams of the four key genes in the C5.0 decision tree model, namely, NFKB1, PHACTR1, PTGS2 and PTX3 genes, wherein a is the ROC curve diagram of the four machine learning models of the benzene exposure group; b is the weight diagram of the four key genes in the C5.0 decision tree model of the benzene exposure group; c is the ROC curve diagram of the four machine learning models in the benzene non-exposure group; and d is the weight diagram of the four key genes in the C5.0 decision tree model in the benzene non-exposure group.

[0041] Fig.10 These are the volcano map and heat map of PTX3 of the present invention, wherein a is the volcano map of PTX3, b is the heat map of PTX3, and c is the GSVA analysis result map of PTX3. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are 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.

[0043] Unless otherwise specified, the test materials or reagents used in the following examples are all existing and can be obtained from commercial channels. If no specific techniques or conditions are specified in the examples, they can be carried out according to conventional techniques or conditions disclosed in the art.

[0044] Abbreviations of various enrichment analyses in the present invention:

[0045] GO: Gene Ontology is a widely used terminology library for describing and counting the relationships between genes and protein functions.

[0046] KEGG: Kyoto Encyclopedia of Genes and Genomes, is a comprehensive database that is roughly divided into three categories: systematic information, genomic information, and chemical information;

[0047] GSEA: Gene Set Enrichment Analysis, a computational biology method for analyzing gene expression data, aims to reveal gene expression patterns associated with specific biological processes, pathways or functions;

[0048] Example

[0049] A method for constructing a benzene-induced blood abnormality assessment model comprises the following steps:

[0050] S1.1: Preparation of benzene damage expression profile datasets: The benzene damage expression profile datasets GSE9569 and GSE21862 were downloaded from the GEO database; the GSE21862 dataset included 83 benzene-exposed workers and 42 unexposed controls exposed to different benzene concentrations; the GSE9569 dataset included 8 benzene-exposed workers and 8 unexposed controls, and the data were compared on two microarray platforms.

[0051] S1.2: Principal component analysis (PCA) was used to eliminate the external differences between the two data sets; the R software "limma" package was used to screen the differentially expressed genes (DEGs) between GSE9569 and GSE21862 with a significance value of P < 0.05 and |logfoldchange (logarithm of the difference) | ≥ 0.5 as the initial screening conditions, and then the DEGs heat map was drawn using the "pheatmap" package and the DEGs volcano map was drawn using the "ggplot2" package.

[0052] S2: The “ClusterProfiler” package in R software was used to perform GO, KEGG and GSEA enrichment analysis on DEGs; GO analysis included three parts: cell components (CC, referring to each part of the cell and the extracellular environment), molecular function (MF, referring to the activity at the molecular level, such as catalytic or binding activity) and biological process (BP, referring to a series of events generated by the orderly combination of one or more molecular functions); p < 0.05 was used as the standard, and bubble charts were used to display the results of enrichment analysis;

[0053] S3.1: Based on the results of enrichment analysis, three machine learning algorithms, namely lasso regression algorithm, SVM-RFE support vector machine recursive feature elimination algorithm and random forest algorithm, were used to screen characteristic genes through R software. Then, the venn package was used to take the intersection of the characteristic genes screened by the three machine learning algorithms to obtain key genes.

[0054] S3.2: Use R software to draw the receiver operating characteristic (ROC) curve of the key genes obtained, and calculate the area under the curve (AUC) value. The larger the AUC value, the better the diagnostic performance. Draw a violin plot for observation and analysis.

[0055] S3.3: Selection of research subjects: A five-year prospective cohort of benzene-exposed workers was established. In the first year, 445 research subjects were selected and divided into a benzene-exposed group and a benzene-non-exposed group, with 214 subjects in the benzene-exposed group and 231 subjects in the benzene-non-exposed group. The key genes and data of the subjects diagnosed with benzene poisoning were obtained through physical examination. The physical examination method was: 4 mL of peripheral blood was drawn from the benzene-exposed group and the benzene-non-exposed group, of which 2 mL was for biochemical testing and 2 mL for routine blood analysis.

[0056] The method for analyzing and determining chronic mild benzene poisoning is as follows: The diagnosis of chronic mild benzene poisoning is based on the "Diagnostic Standards for Occupational Benzene Poisoning" GBZ 68-2022: There is a history of close occupational contact with benzene for 3 months or more, which may be accompanied by clinical manifestations such as dizziness, headache, fatigue, insomnia, memory loss, and repeated infections. Peripheral blood cell analysis is reviewed every 2 weeks within 3 months, and one of the following conditions is met: a) White blood cell count is less than 3.5x10 9 / L(normal range 3.5x10 9 -9.5x10 9 / L); b) Neutrophil count is less than 1.8x10 9 / L(normal range 1.8x10 9 -6.3x10 9 / L); c) Platelet count is less than 80x109 / L (normal range 125x10 9 -350x10 9 / L).

[0057] S3.4: SPSS 27.0 software was used for data analysis. For data that conformed to normal distribution, The data that did not conform to the normal distribution were expressed as medians and quartiles, and the differences between the groups were compared using non-parametric tests; p < 0.05 was considered to be significantly different;

[0058] The “GSVA” package was used to evaluate genomic-related pathways. GSVA: Gene set variation analysis is a nonparametric, unsupervised analysis method that is mainly used to evaluate the gene set enrichment results of sequencing.

[0059] After population verification and statistical analysis of the correspondence between key genes and benzene poisoning, the benzene damage marker genes were obtained.

[0060] S4.1: Train and test multiple learning models based on the obtained benzene damage marker genes to build a machine learning model:

[0061] The data set is divided into a training group and a test group in a ratio of 7:3. First, the taboo search algorithm in the R4.1.1bnlearn package is used to learn the data and generate the model network structure. Then, the model structure is adjusted and optimized in combination with the literature to form the final model framework. The next step is to calculate the four model parameters. This study uses Netica6.09 software to build the final model, and the parameters of the model are formed by learning the data. The following training and testing learning models are C5.0DT, SVM, Bayes, BP, the training, testing and construction methods of each model are as follows in Table 1-4:

[0062] Table 1 Construction method of C5.0DT model

[0063] Parameter name Parameter Value Training time 0.207s Data segmentation 0.7 Data shuffle yes Cross Validation 10 Activation Function identity Solver lbfgs Learning Rate 0.1 L2 regularization term 1 Iterations 1000 Number of neurons in the first hidden layer 100

[0064] Table 2 SVM model construction method

[0065] Parameter name Parameter Value Training time 0.043s Data segmentation 0.7 Data shuffle yes Cross Validation 10 Penalty coefficient 1 Kernel Function linear Kernel function coefficients scale Kernel function constant 0 The highest order of kernel function 3 Error convergence condition 0.001 Maximum number of iterations 1000 Multi-classification fusion strategy ovr

[0066] Table 3 Bayesian construction method

[0067] Parameter name Parameter Value Training time 0.021s Data segmentation 0.7 Data shuffle yes Cross Validation 10 Prior distribution Gaussian distribution alpha 1 Binarization threshold 0

[0068] Table 4 Construction method of BP model

[0069] Parameter name Parameter Value Training time 1.048s Data segmentation 0.7 Data shuffle yes Cross Validation 10 Node Split Evaluation Criteria gini Number of decision trees 100 Sampling with replacement true Out-of-bag data testing false The maximum feature ratio considered when splitting auto Minimum number of samples for internal node splitting 2 Minimum number of samples for leaf nodes 1 The minimum weight of samples in a leaf node 0 Maximum depth of the tree 10 Maximum number of leaf nodes 50 Node partition impurity threshold 0

[0070] S4.2: Use the test set to first train the support vector machine model, Bp neural network model, Bayesian model and C5.0 decision tree, and then use the training set to test them; obtain the confusion matrix of different models, calculate the test result accuracy of each learning model and draw the ROC curve, and compare them. Use GraphPad Prism 8.3 software to draw the comparison results, and select the optimal model as the evaluation model for benzene-induced blood abnormalities.

[0071] Result analysis:

[0072] 1. Genetic analysis results:

[0073] After eliminating the external differences between the two data sets through PCA analysis (Principal Component Analysis) Figure 1 a, b), as can be seen from the figure, there is a large overlap between the two data sets after adjustment; a total of 40 DEGs (differentially expressed genes) were screened out by differential analysis, of which 38 were up-regulated genes and 2 were down-regulated genes, and volcano plots and heat maps were drawn. Volcano plot ( Figure 1 c) shows the differentially expressed genes between GSE9569 and GSE21862, with red representing up-regulated genes and green representing down-regulated genes; cluster heat map ( Figure 1 d) shows the specific expression of DEGs in each sample.

[0074] GO, KEGG, and GSEA enrichment analysis were performed on the screened differentially expressed genes: GO enrichment analysis found that ( Figure 2 a) Biological processes (BP) mainly involve a series of inflammatory response processes such as cytokine-mediated signal transduction pathways, leukocyte migration, leukocyte-cell adhesion and response to lipopolysaccharide; cellular components (CC) mainly involve tertiary granules, specific granules, serine peptidase complexes, etc.; molecular functions (MF) mainly involve cytokine receptor binding, cytokine activity, receptor ligand activity, etc. KEGG enrichment analysis results ( Figure 2 b) showed that the enriched pathways were mainly concentrated in the inflammation and immune response pathways, such as cytokine-cytokine receptor interaction, interaction between viral protein and cytokine and cytokine receptor, chemokine signaling pathway and NF-KB signaling pathway. Further GSEA enrichment analysis revealed that DEGs were enriched in 5 positive gene sets in the high expression group, mainly involving cytokine receptor interaction, graft-versus-host disease, jak stat signaling pathway, leishmaniasis infection and node receptor-like signaling pathway ( Figure 2 c); Five positive gene sets were enriched in the low expression group, mainly involving base excision repair, Huntington's disease, oxidative phosphorylation, Parkinson's disease and RNA degradation ( Figure 2 d).

[0075] 2. Verification results of key genes:

[0076] The LASSO algorithm was used to screen 26 characteristic genes from DEGs ( Figure 3 a, b), 29 characteristic genes were screened using the SVM-RFE algorithm ( Figure 3 c, d), 11 characteristic genes were screened out using the RF algorithm ( Figure 3 e,f), and four genes were identified after the intersection of the three (Figure g), namely: NFKB1, PHACTR1, PTGS2 and PTX3. Then, ROC curves were drawn for the four genes respectively ( Figure 4 ), and their AUCs (95% CI) were 0.822 (0.743-0.893), 0.755 (0.680-0.824), 0.802 (0.714-0.873) and 0.834 (0.756-0.901), respectively. GEO data were extracted to draw violin plots for the four key genes ( Figure 5 ), among which four genes showed large differences between the two groups.

[0077] 3. Population verification and statistical analysis results:

[0078] The basic population analysis was conducted on the benzene-exposed group and the benzene-non-exposed group. The basic characteristics of the subjects are shown in Table 5. The statistical comparison results of blood routine tests and the four genes are shown in Figure 6 ; After comparison, we found that the p values ​​of 8-OHdG, MDA, Neutrophi, PLT, S-PMA, WBC, NFKB1 and PTX3 were all less than 0.05, which was statistically significant; while the p values ​​of PHACTR1 and PTGS2 were both greater than 0.05, which was not statistically significant; therefore, it can be considered that compared with the non-benzene exposed group, the changes in 8-OHdG, MDA, Neutrophi, PLT, S-PMA, WBC, NFKB1 and PTX3 in the benzene exposed group were obvious, which may serve as markers of benzene damage, while the changes in PHACTR1 and PTGS2 in the benzene exposed group cannot be considered significant, that is, PHACTR1 and PTGS2 cannot be identified as markers of damage.

[0079] Subsequently, according to GBZ 68-2022 "Diagnostic Criteria for Occupational Benzene Poisoning", the benzene exposure group was divided into a blood abnormality group and a blood normal group, and statistical analysis was performed on the four key genes (see Table 6). The P value of PTGS2 was 0.136>0.05, which was not statistically significant. The statistical comparison results of each factor were plotted, as shown in Figure 7As shown in the results, it can be seen that the p values ​​of PTGS2 and PLT are greater than 0.05, which is not statistically significant; while the other p values ​​are less than 0.05, which is statistically significant. It can be considered that compared with the normal blood group, the changes in 8-OHdG, MDA, Neutrophi, S-PMA, WBC, NFKB1, PHACTR1 and PTX3 in the abnormal blood group are obvious, but it cannot be considered that the changes in PTGS2 and PLT in the abnormal blood group are obvious.

[0080] Table 5. Basic characteristics of the 445 subjects

[0081]

[0082]

[0083] The superscript a indicates that a nonparametric rank sum test is used.

[0084] Table 6. Statistical analysis of four genes in the abnormal blood group and the normal blood group

[0085] Blood abnormality group (n=21) Normal blood group (n=193) p-value <![CDATA[NFKB1 a ]]> 4.789(2.961-6.038) 2.215(1.659-3.725) ＜0.001 <![CDATA[PHACTR1 a ]]> 2.075(0.848-2.830) 0.970(0.219-1.899) 0.004 <![CDATA[PTGS2 b ]]> 0.997±0.501 1.237±0.714 0.136 <![CDATA[PTX3 a ]]> 1.903(1.577-2.140) 2.272(2.003-2.847) ＜0.001

[0086] The superscript a indicates that the nonparametric rank sum test was used; the superscript b indicates that the independent sample t test was used;

[0087] 4. Determination of machine algorithm:

[0088] GraphPad Prism 8.3 software was used to plot the results obtained by different prediction models in the training set and test set. Figure 8 The accuracy and AUC values ​​are used to measure the model performance, and the ROC curve is combined with the ROC curve ( Fig. 9 ) to compare the four prediction models, among which the AUC values ​​of the C5.0 decision tree in the benzene exposure group and the benzene non-exposure group were the largest (0.823, 0.959), indicating that the C5.0 decision tree had the best diagnostic performance. Therefore, the C5.0 decision tree can make a better prediction of benzene damage, and then the weight diagram of the C5.0 decision tree model was drawn ( Fig. 9 b, d), according to the figure, PTX3 is the most important gene. Then draw the volcano map and heat map of PTX3 ( Fig.10 a,b).

[0089] 5. Gene set variation analysis results

[0090] The relevant pathways of PTX3 in different risk groups were further explored. Fig.10c shows that in the up-regulated group, PTX3 triggered valine and isoleucine degradation, glycine serine and threonine metabolism, and the interconversion of pentose and glucuronic acid, etc.; in contrast, PTX3 in the down-regulated group triggered cytokine receptor interaction, nodal receptor-like signaling pathway, and jak-stat signaling pathway, etc.

[0091] Conclusion: In summary, the present invention combines bioinformatics and machine learning to explore the key risk genes in damage caused by benzene exposure, and finds that the two genes NFKB1 and PTX3 may be involved in the blood toxicity of benzene through inflammatory response. Combining the key risk genes verified by the present invention and the machine learning algorithm, the risk of benzene poisoning can be accurately assessed and patients with early symptoms can be identified.

[0092] Based on the above conclusions, the present invention believes that NFKB1 and PTX3 can be used as marker genes or markers for benzene-induced blood abnormalities, that is, the presence of benzene-induced blood abnormalities can be determined by judging the gene abnormalities or expression abnormalities of NFKB1 and PTX3.

[0093] Meanwhile, the present invention proposes the use of primers for detecting NFKB1 and / or PTX3 in the preparation of reagents for detecting benzene-induced blood abnormalities.

[0094] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for constructing a benzene-induced blood abnormality assessment model, characterized in that: The steps include: S1: First, principal component analysis was performed on the benzene damage expression profile dataset, and then the initial screening conditions were set to screen out the differentially expressed genes between the benzene damage expression profile datasets; S2: Perform GO, KEGG, and GSEA enrichment analysis on the differentially expressed genes to obtain the analysis results; S3: Use different machine learning algorithms to screen out characteristic genes from the analysis results and take the intersection to obtain key genes, then perform verification analysis on the key genes to obtain benzene damage marker genes; S4: Use benzene damage marker genes to train and test multiple learning models, and select a suitable learning model as a benzene-induced blood abnormality assessment model based on the test results of the learning model.

2. The method for constructing a benzene-induced blood abnormality assessment model according to claim 1, characterized in that: The initial screening conditions in S1 include: using the R software "limma" package, with the significance value P<0.05 and |logfoldchange|≥0.5 as the initial screening conditions; after screening the differentially expressed genes, using the "pheatmap" package to draw a heat map, and using the "ggplot2" package to draw a volcano map.

3. The method for constructing a benzene-induced blood abnormality assessment model according to claim 2, characterized in that: The step S2 includes using the "ClusterProfiler" package in R software to perform GO, KEGG, and GSEA enrichment analysis on the differentially expressed genes, with a significance value of p < 0.05 as the standard, and using a bubble chart to display the enrichment analysis results.

4. The method for constructing a benzene-induced blood abnormality assessment model according to claim 3, characterized in that: The machine learning algorithms in S3 include: lasso regression algorithm, SVM-RFE support vector machine recursive feature elimination algorithm and random forest algorithm.

5. The method for constructing a benzene-induced blood abnormality assessment model according to claim 4, characterized in that: The steps of S3 include: S3.1: Three machine learning algorithms, namely, lasso regression algorithm, SVM-RFE support vector machine recursive feature elimination algorithm and random forest algorithm, were used to screen characteristic genes through R software. Then, the venn package was used to take the intersection of the characteristic genes screened by the three machine learning algorithms to obtain key genes. S3.2: Use R software to draw the receiver operating curve for the obtained key genes, calculate the area under the curve (AUC) value, and the larger the AUC value, the better the diagnostic performance; and draw a violin plot; S3.3: Select research subjects, conduct population verification and statistical analysis on the correspondence between key genes and benzene poisoning, and obtain benzene damage marker genes.

6. The method for constructing a benzene-induced blood abnormality assessment model according to claim 5, characterized in that: The multiple learning models in S4 include: support vector machine model, Bp neural network model, Bayesian model and C5.0 decision tree; the method of selecting a suitable learning model according to the test results of the learning model includes: calculating the accuracy of the test results of each learning model and drawing the ROC curve, and comparing them, and using GraphPad Prism 8.3 software to draw the comparison results to obtain a suitable learning model; preferably, the trained C5.0 decision tree is used as the benzene-induced blood abnormality assessment model.

7. A benzene-induced blood abnormality assessment model obtained by the construction method according to any one of claims 1 to 6.

8. Use of the construction method according to any one of claims 1 to 6 or the benzene-induced blood abnormality assessment model according to claim 7 in a blood abnormality detection model in a benzene exposure environment.

9. A benzene-induced blood abnormality marker obtained by screening using the benzene-induced blood abnormality assessment model as claimed in claim 7, characterized in that: Including NFKB1 and PTX3.

10. Use of primers for detecting NFKB1 and / or PTX3 in the preparation of reagents for detecting benzene-induced blood abnormalities.

Images