中国肿瘤外科杂志2026，Vol.18Issue(1)：36-45,10.DOI:10.3969/j.issn.1674-4136.2026.01.006
基于机器学习的结直肠癌内质网应激相关基因识别与分析
Identification and analysis of endoplasmic reticulum stressrelated genes in colorectal cancer using machine learning
摘要
Abstract
Objective To investigate the expression characteristics and clinical significance of endoplasmic reticulum stress-related genes in colorectal cancer(CRC)and evaluate their potential application in diagnosis and prognosis.Methods This study selected three CRC datasets(GSE41258,GSE50760,and GSE5206)from the GEO database.After data preprocessing,differentially expressed genes(DEGs)were identified through differential expression analysis.The DEGs were intersected with endoplasmic reticulum stress-related genes,and key genes were identified using two machine learning methods:support vector machine-recursive feature elimination(SVM-RFE)and gradient boosted decision tree(GBDT).Seven intersecting genes(HSD11B2,GCG,RCN1,COL1A1,TRIB3,CCND1,and VEGFA)were ultimately selected.Based on the expression values of these genes,an artificial neural network(ANN)model was constructed to evaluate diagnostic performance,and its generalizability was assessed using ROC curve analysis in the combined dataset and individual datasets.The expression differences and correlations of key genes between cancer andnormal tissues were analyzed.Survival analysis was conducted using the TCGA database to explore the relationship between gene expression levels and overall survival(OS)in patients.Finally,potential drugs targeting these key genes were identified using the Enrichr database.Results The seven identified key genes exhibited significant expression differences in CRC tissues.GCG and HSD11B2 were highly expressed innormal tissues,while RCN1,COL1A1,TRIB3,CCND1,and VEGFA were highly expressed in cancer tissues.The ANN model achieved an AUC greater than 0.8 in both the training and validation groups,indicating good diagnostic performance.Survival analysis showed that high expression of GCG and HSD11B2 was associated with better prognosis,whereas high expression of the other genes was associated with poor prognosis.Drug screening identified potential agents such as Entinostat and Indomethacin that may regulate the expression of these genes.Conclusions The expression of HSD11B2,GCG,RCN1,COL1A1,TRIB3,CCND1,and VEGFA has significant diagnostic and prognostic value in CRC and may provide new directions for targeted therapy research.关键词
结直肠癌/内质网应激/机器学习/生物信息学分析/药物筛选Key words
Colorectal cancer/Endoplasmic reticulum stress/Machine learning/Bioinformatics analysis/Drug screening引用本文复制引用
潘荣天,张圆,于韶荣..基于机器学习的结直肠癌内质网应激相关基因识别与分析[J].中国肿瘤外科杂志,2026,18(1):36-45,10.基金项目
国家自然科学基金(82172872,81902489) （82172872,81902489）
江苏省肿瘤医院移山计划项目(YSPY202408) （YSPY202408）
江苏省重点研发计划项目(BE2021745) （BE2021745）
江苏省自然科学基金(BK20191079) （BK20191079）
