摘要
Abstract
Objective To construct and validate a prognostic model for glioma based on perineural invasion(PNI)-re-lated features,and to evaluate its potential for identifying chemoresistant subpopulations through exploratory analysis.Methods Transcriptomic and clinical data from patients treated with temozolomide(TMZ)were retrieved from the Chi-nese Glioma Genome Atlas(CGGA).Differentially expressed genes between primary and recurrent tumors were intersect-ed with PNI-related gene sets,and the resulting overlapping genes were subjected to functional enrichment analysis.A comprehensive framework integrating 101 machine learning algorithm combinations was applied for feature selection and risk score construction.Predictive performance was assessed using survival analysis,time-dependent receiver operating characteristic(ROC)curves,and nomograms.Drug sensitivity prediction was performed as an exploratory secondary analy-sis to compare the half-maximal inhibitory concentration(IC₅₀)between risk subgroups.Results Sixty-five core genes asso-ciated with PNI and recurrence were identified.The StepCox+RSF algorithm,which exhibited optimal performance,was used to establish a 5-gene(CP,SNAI2,GLP1R,LRP1B,and VCAN)prognostic model.Patients in the high-risk group had significantly shorter overall survival(P<0.000 1),and the model demonstrated robust and consistent predictive performance at 1,3,and 5 years across cohorts.A nomogram integrating clinical parameters,including IDH status,further improved pre-dictive accuracy.The high-risk group exhibited significantly higher predicted IC₅₀ values for carmustine(P=1.3×10⁻¹⁰)and multiple targeted agents,suggesting potentially lower drug sensitivity.Conclusion The 5-gene prognostic model,devel-oped via integrated machine learning approaches,enables reliable survival risk stratification in glioma patients.Exploratory drug sensitivity analysis indicates that this model may facilitate the identification of high-risk subgroups with a propensity for reduced chemosensitivity.关键词
脑胶质瘤/周围神经浸润/机器学习/生物信息学分析/预后模型/化疗耐药/药物敏感性预测Key words
Glioma/Perineural invasion/Machine learning/Bioinformatics analysis/Prognostic model/Chemoresis-tance/Drug sensitivity prediction分类
医药卫生
