摘要
目的:建立基于GEO数据库硬皮病溶酶体相关基因的随机森林和人工神经网络(artificial neural network,ANN)联合诊断模型并评价其效果。方法:通过GEO数据库获取4份硬皮病芯片,从AmiGO2数据库中获取875个溶酶体相关基因。其中GSE95065及GSE76807合并作为实验组数据集,使用随机森林算法筛选硬皮病溶酶体相关特征基因,并用特征基因构建人工神经网络模型,用10折交叉验证模型准确性。再用验证数据集GSE32413与GSE59787对模型进一步验证,利用ROC曲线下面积值评估模型准确性。最后用实时荧光定量PCR(real-time quantitative PCR,RT-qPCR)进行实验验证。结果:共获取差异基因46个,其中上调基因16个,下调基因30个。进一步通过随机森林筛选得到最相关的6个特征基因(LYN、TNFAIP3、RNF128、MCOLN3、ANKFY1、PLD3),并构建ANN诊断模型。使用该模型绘制了实验组和验证组诊断的ROC曲线,AUC值为0.999。10折交叉验证AUC平均值大于0.980。验证组AUC为0.740和0.732。RT-qPCR结果表明与对照组相比,硬皮病中LYN(P=0.004)、TNFAIP3(P=0.0001)表达量明显上调,而RNF128(P=0.0002)、MCOLN3(P=0.001)、ANKFY1(P=0.02)、PLD3(P<0.0001)表达量在硬皮病组中明显下调。与机器学习算法结果相一致。结论:构建了硬皮病溶酶体相关特征基因的ANN诊断模型,为探索硬皮病发病机制提供了一个新视角。
Objective:To establish a combined random forest and artificial neural network diagnosis model of lysosome related genes in scleroderma based on GEO database and evaluate its effect.Method:Four scleroderma chips were obtained from GEO database.GSE95065 and GSE76807 were combined as training data set,and random forest algorithm was used to screen scleroderma lysosome‑related characteristic genes.The artificial neural network model was constructed with characteristic genes,and the accuracy of the model was verified by 10‑fold crossover.Then the verification data set GSE32413 and GSE59787 are used to further verify the model,and the product value under the ROC curve is used to evaluate the accuracy of the model.Finally,RT‑qPCR was used for experimental verification.Result:The results showed a total of 46 differentially expressed genes were identified,including 16 genes that were up‑regulated and 30 genes that were down‑regulated.Furthermore,the six most relevant characteristic genes(LYN,TNFAIP3,RNF128,MCOLN3,ANKFY1,PLD3)were screened by random forest,and the artificial neural network diagnosis model was constructed.Using this model,the ROC curves of training group and verification group were drawn,and the AUC value was 0.999.The AUC of the verification group was 0.740 and 0.732,respectively.The average AUC of 10%discount cross‑validation is greater than 0.980.RT‑qPCR results showed that compared with the control group,the expressions of LYN(P=0.004)and TNFAIP3(P=0.0001)were significantly up‑regulated in scleroderma,while the expressions of RNF128(P=0.0002),MCOLN3(P=0.001),ANKFY1(P=0.02)and PLD3(P<0.0001)were significantly down‑regulated in scleroderma.Consistent with machine learning results.Conclusion:An artificial neural network diagnosis model of lysosome‑related characteristic genes in scleroderma was constructed,which provides a new perspective for exploring the pathogenesis of scleroderma.
作者
左志威
卞博
崔家康
耿玉鑫
王一晨
郭克磊
孟庆良
卞华
ZUO Zhiwei;BIAN Bo;CUI Jiakang;GENG Yuxin;WANG Yichen;GUO Kelei;MENG Qingliang;BIAN Hua(School of Orthopedics and Traumatology,Henan University of Traditional Chinese Medicine,Department of Rheumatology,Henan Provincial Hospital of Traditional Chinese Medicine,Zhengzhou 450008,China;College of Traditional Chinese Medicine,North China University of Science and Technology,Tangshan 063210,China;Henan Key Laboratory of Zhang Zhongjing Formulae and Herbs for Immunoregulation,Nanyang Institute of Technology,Nanyang 473004,China)
出处
《海南医科大学学报》
北大核心
2025年第2期109-117,共9页
Journal of Hainan Medical University
基金
国家自然科学基金项目(82074415)
中原英才计划-中原科技创新领军人才项目(234200510006)
河南省科技计划项目(232102311201)
南阳市基础与前沿技术研究专项计划重点项目(23JCQY1006)。
关键词
系统性硬化症
溶酶体
人工神经网络
随机森林
诊断模型
Systemic sclerosis
Lysosome
Artificial neural network
Random forest
Diagnostic model
