Mining The GEO Database For The Effects Of DMARDs On Rheumatoid Arthritis And Identifying Herbal Medicines

ObjectiveBioinformatics and machine learning approach were used to identify differentially expressed genes associated with disease development in rheumatoid arthritis(RA)synovium,to investigate the effects of DMARDs such as methotrexate,tolimumab and rituximab on RA synovial differentially expressed genes(DEGs)and to explore the immune infiltration of the synovium for drug prediction.MethodsFrom the GEO(Gene Expression Omnibus)database,three microarray data files GSE12021,GSE55457,and GSE77298 of human rheumatoid arthritis synovial tissue gene expression were obtained as the training set,and microarray files GSE7307,GSE55235,and GSE89408 were downloaded as the validation set,while obtained the synovial expression profile microarrays GSE45867,GSE24742,GSE97165 before and after RA treatment.The three sets of data were combined to remove the batch effect.DEGs in the training and treatment groups were screened using R software.Venn diagrams of differentially expressed genes between RA and healthy controls before and after treatment were produced,and analyzed for GO function enrichment and KEGG pathway enrichment.Candidate diagnostic gene sets were screened from the training group by support vector machine recursive feature elimination(SVM-RFE),LASSO and random forest(RF),respectively.The expression levels of overlapping genes of the three algorithms were further validated in the validation set.ROC curves were generated using the training set expression data.The area under the ROC curve(AUC)was used to determine the diagnostic validity of RA and control samples.GSEA analysis and PPI network analysis were also performed,while the relative proportion of immune cell infiltration was calculated(CIBERSORT algorithm).Spearman’s rank correlation analysis in R software was used to explore the relationship between the identified genetic markers and the level of infiltrating immune cells.Finally,potential compounds in the HERB database for RA were screened using the TCM Systematic Pharmacology Database and Analysis Platform(TCMSP),and the Auto Dock tool was used to construct molecular docking models between drugs and target genes.ResultsCompared with HC,there were 1173 upregulated and 628 downregulated DEGs in RA synovial tissue,which were mainly enriched in "immune response regulatory signaling pathways" and "leukocyte-mediated immunity".20 DEGs were significantly upregulated and 30 DEGs downregulated in the methotrexate treatment group;100DEGs were significantly upregulated and 55 DEGs downregulated in the rituximab treatment group;91 DEGs were significantly upregulated and 317 DEGs downregulated in the tolimumab treatment group.These DEGs were enriched in the regulatory cell adhesion,leukocyte-cell adhesion,leukocyte transfer,and insulin-like growth factor receptor signaling pathways.Notably,a total of 296 highly expressed DEGs were down-regulated and 27 low-expressed DEGs were up-regulated after treatment.Six genetic diagnostic biomarkers were screened from 323 genes using three machine learning algorithms: AIM2,CXCL13,FLVCR2,MICB,OAS2,SLAMF8 as characteristic genes,their identification was validated in the validation set and PPI network was constructed.There were 10 immune cells with significantly different expression in the immune cell composition between RA and control synovial tissue.One natural compound(quercetin),the active ingredient of 188 herbs,was screened based on diagnostic genes.Target-compound-herb network was constructed,and the molecular docking models of quercetin-FLVCR2,quercetin-MICB were constructed using Auto Dock tool.ConclusionAIM2,CXCL13,FLVCR2,MICB,OAS2,SLAMF8 could be used as characteristic genes of RA synovium.Ten immune cells such as memory B cells and plasma cells may be involved in the immune infiltration of RA synovial membrane.Quercetin may be a potential therapeutic agent for rheumatoid arthritis targeting FLVCR2 and MICB.