人工神经网络筛选类风湿关节炎的特征基因及免疫细胞浸润分析

Artificial neural network for screening characteristic genes of rheumatoid arthritis and analysis of immune cell infiltration

目的:运用人工神经网络确定类风湿关节炎(RA)的特征基因,并分析免疫细胞在RA相关微环境中的作用。方法:GSE1919和GSE77298芯片均来自GEO数据库。运用R语言将两芯片进行合并与批次矫正,得到一个新数据集,并进行差异分析。对差异表达基因(DEGs)进行Metascape富集分析和GO、KEGG富集分析。运用R软件“randomForest”包在随机森林(RF)算法下筛选RA特征基因,并依据基因评分构建人工神经网络模型。提取排名前4的基因(HubGene)进行后续分析。应用单样本基因集富集分析(ssGSEA)计算样品中免疫细胞丰度并进行相关性分析。结果:人工神经网络模型筛选出排名前15的基因作为RA特征基因:STAT1、RUNX3、AR、CDH11、LMO4、TIMP1、PLXNC1、CAP2、PRKAA2、VDR、SPP1、HCK、EPHB2、KCNAB1、ITGB7,其中STAT1、RUNX3、CDH11在RA滑膜组织中均上调,AR在RA滑膜组织中下调。免疫细胞浸润结果显示,RA与活化CD4 T细胞相关性最显著。与RUNX3存在显著相关性的免疫细胞最多。RUNX3与活化B细胞、活化CD4 T细胞、活化CD8 T细胞、中央记忆CD4+T细胞、效应性记忆CD8 T细胞、调节性T细胞、γδT细胞和巨噬细胞呈显著正相关,但与NK细胞呈显著负相关。结论:通过人工神经网络确定了与RA相关的15个特征基因,其中排名前4的基因为STAT1、RUNX3、AR、CDH11。强调了活化CD4 T细胞、调节性T细胞、γδT细胞、巨噬细胞、NK细胞、活化B细胞等免疫细胞在RA发病机制中的重要性,为RA诊断和免疫细胞分子机制研究提供了新见解。

Objective:To identify characteristic genes of rheumatoid arthritis(RA)by artificial neural network and to analyze role of immune cells in RA related microenvironment.Methods:GSE1919 and GSE77298 chips were from GEO database.Two chips were combined and batch corrected using R language to obtain a new data set,and difference was analyzed.Metascape enrichment analysis and GO and KEGG enrichment analysis were conducted for differential expressed genes(DEGs)."randomForest"package in R software was used to screen characteristic genes of RA under random forest(RF)algorithm,and artificial neural network model was constructed according to gene score.Top 4 genes(HubGene)were extracted for subsequent analysis.Single sample gene set enrichment analysis(ssGSEA)was used to calculate abundance of immune cells in samples and carry out a series of correlation analysis.Results:Top 15 genes were selected as characteristic genes of RA through artificial neural network model:STAT1,RUNX3,AR,CDH11,LMO4,TIMP1,PLXNC1,CAP2,PRKAA2,VDR,SPP1,HCK,EPHB2,KCNAB1,ITGB7.STAT1,RUNX3 and CDH11 were up-regulated in RA synovial tissue,while AR was down-regulated in RA synovial tissue.Immune cell infiltration results showed that RA had the most significant correlation with activated CD4 T cells.Number of immune cells significantly related to RUNX3 was the largest.RUNX3 was significant positive correlated with activated B cells,activated CD4 T cells,activated CD8 T cells,central memory CD4+T cells,effector memory CD8 T cells,regulatory T cells,γδT cells and macrophages,while significant negative correlated with NK cells.Conclusion:Fifteen characteristic genes related to RA are identified through artificial neural network,among which STAT1,RUNX3,AR and CDH11 are top 4 genes.It emphasizes activation of CD4 T cells,regulatory T cells,γδT cells,macrophages,NK cells,activated B cells and other immune cells are important in pathogenesis of RA,providing new insights for diagnosis of RA and study of molecular mechanism of immune cells.