基于生物信息学分析类风湿关节炎铜死亡相关基因及中药筛选预测

Analysis of Cuproptosis-associated Genes in Rheumatoid Arthritis Based on Bioinformatics and Screening Prediction of Traditional Chinese Medicine

目的 利用生物信息学研究类风湿关节炎(Rheumatoid arthritis, RA)铜死亡相关分子模式和诊断性生物标志物,同时预测具有潜在治疗作用的中药。方法 从基因表达综合数据库中下载GSE55235、GSE55457、GSE77298表达谱数据作为训练数据集,提取铜死亡相关基因进行分析。选择RA患者和健康对照组之间差异表达的铜死亡相关基因(Differentially expressed cuproptosis-associated gene, DECAG),分析DECAG相关的免疫浸润、生物学功能。根据DECAG的表达量对RA患者进行共识聚类分型,基于分型进行加权基因共表达网络分析(Weighted gene co-expression network analysis, WGCNA)来识别核心基因。取degree值前100的核心基因做GO和KEGG富集分析并构建训练模型,训练模型包括随机森林模型(Random forest, RF)和支持向量机模型(Support vector machine, SVM)、极端梯度提升(eXtreme gradient boosting, XGB)模型和广义线性模型(Generalized linear model, GLM),筛选与RA特征最相关的5个基因作为诊断性生物标志物并进行验证。最后进行中药预测。结果 获得6个DECAG(NLRP3、SLC31A1、LIAS、CDKN2A、DBT、DLST)。使用共识聚类方法将RA基因分为两个亚型(C1,C2),分型WGCNA获得核心基因418个。取degree值前100的核心基因构建训练模型,基于对|残差|的反向累积分布图和|残差|的箱线图的分析,发现SVM模型与其他3种模型相比,SVM模型维持最低的|残差|分布。总体受试者工作特征曲线(Receiver operating characteristic curve, ROC)结果显示,SVM模型比其他3个模型具有更高的网络曲线下面积(Area under the curve, AUC)值(AUC:0.966)。综合考虑,SVM模型为最合适的训练模型。获得5个与RA特征最相关的基因(TMOD3、PIK3CG、WASL、FGF4、GSN),基于5个RA特征基因的表达水平,建立了临床应用列线图,并且决策曲线分析(Decision curve analysis, DCA)图和校正曲线图也显示结果具有良好的预测准确度。中药预测结果显示郁金、莪术、红花最有可能具有治疗RA的作用。结论 铜死亡在RA的发生和诊断中起重要作用。

Objective To study the molecular patterns and diagnostic biomarkers of cuproptosis-associated in rheumatoid arthritis(RA)by bioinformatics and predict the potential therapeutic effect of traditional Chinese medicine.Methods The expression profiles of GSE55235,GSE55457 and GSE77298 were downloaded from Gene Expression Omnibus database as the training set.Cuproptosis-associated related genes were extracted for analysis.The differentially expressed cuproptosis-associated gene(DECAG)between RA patients and healthy controls was selected.DECAG related immune infiltration and biological function were analyzed.RA patients were classified by consensus clustering according to the expression of DECAG.Weighted gene co-expression network analysis(WGCNA)was constructed based on genotyping to identify core modules and core genes.GO analysis and KEGG analysis were performed on the top 100 core genes of degree value and the training model was constructed,including Random Forest(RF)and Support Vector Machine(SVM),eXtreme Gradient Boosting(XGB)and Generalized Linear Model(GLM).Five genes most related to RA characteristics were screened and verified as diagnostic biomarkers.Finally,Chinese medicine prediction was carried out.Results SixDECAGs(NLRP3,SLC31A1,LIAS,CDKN2A,DBT,DLST)were obtained.RA genes were classified into two isoforms(C1,C2)using a consensus clustering method,and 418 core genes were obtained by typing WGCNA.The machine learning model was constructed by taking the top 100 core genes of degree value.Based on the analysis of the reverse cumulative distribution map|residual|and the box plot of|residual|,it was found that the SVM model maintaineds the lowest|residual|distribution compared with the other three models.From the overall Receiver operating characteristic curve(ROC)analysis,the SVM model had a higher area under the curve(AUC)value(AUC:0.966)than the other three models.Taken together,SVM mode was the most appropriate training model.Five genes(TMOD3,PIK3CG,WASL,FGF4,GSN)mostly related to RA characteristics were obtained.Based on the expression levels of five RA characteristic genes,a clinical application nomogram was established,and the decision curve analysis(DCA)diagram and correction curve diagram also showed good prediction accuracy.Herbal predictions showed that Yujin(Curcumae Radix),Ezhu(Curcumae Rhizoma)and Honghua(Carthami Flos)were most likely to have therapeutic effects in RA.Conclusion Cuproptosis-associated plays an important role in the occurrence and diagnosis of RA.