机器人外科学杂志（中英文）2026，Vol.7Issue(5)：912-919,8.DOI:10.12180/j.issn.2096-7721.2026.05.027
基于机器学习算法筛选直肠癌新辅助放化疗敏感相关基因
Machine learning-based screening of genes associated with sensitivity to neoadjuvant chemoradiotherapy in rectal cancer
摘要
Abstract
Objective:To integrate multiple rectal cancer transcriptome datasets and screen genes associated with sensitivity to neoadjuvant chemoradiotherapy using machine learning algorithms,aiming to identify potential biomarkers and therapeutic targets.Methods:Five datasets related to neoadjuvant concurrent chemoradiotherapy for rectal cancer were systematically retrieved and integrated from the GEO database.After excluding samples with incomplete clinical information,a total of 723 patients were included,of which 453 were sensitive to neoadjuvant concurrent chemoradiotherapy(sensitive group)and 270 were resistant(resistant group).After removing batch effects,differential expression analysis was performed.11 machine learning algorithms,including Random Forest,Support Vector Machine,Generalized Linear Model,Gradient Boosting Machine,K-Nearest Neighbors,Neural Network,L1-regularized Logistic Regression,Naive Bayes,Adaptive Boosting,Decision Tree,and Bagging,were used to screen genes associated with radiosensitivity.Important genes were identified through protein-protein interaction network analysis,and hub genes were further identified using four topological algorithms(Degree,EPC,MCC,MNC)implemented in Cytoscape software.Single-cell transcriptome databases were integrated to deeply characterize the cellular origin and functional features of the hub genes.Results:Among the 11 machine learning algorithms,except for the Generalized Linear Model and Decision Tree,the area under the precision-recall curve(AUPRC)of the remaining algorithms was>0.7,with the Support Vector Machine performing best,reaching 0.8241,indicating that machine learning algorithms have excellent discriminative ability in high-dimensional gene expression data.Through protein-protein interaction network and topological algorithms,seven genes associated with radiosensitivity were identified:IFNG,KLRD1,CD40,CD2,CRTAM,TBK1,and DHDH,with IFNG recognized as the hub gene.Single-cell transcriptome analysis indicated that IFNG was specifically enriched in CD8+T cell and natural killer cell subsets,establishing its role as a key effector molecule in antitumor immunity.This molecular signature is highly consistent with the favorable efficacy achieved in recent phase Ⅲ clinical trials of short-course radiotherapy combined with immunotherapy in neoadjuvant therapy for rectal cancer,further supporting the important value of synergistic interaction between immunotherapy and chemoradiotherapy in the neoadjuvant treatment of rectal cancer.Conclusion:This study identified a robust set of immune-related gene biomarkers associated with chemoradiotherapy sensitivity and revealed the crucial role of IFNG-mediated cytotoxic immune infiltration in neoadjuvant therapy for rectal cancer,providing a basis for individualized neoadjuvant chemoradiotherapy in rectal cancer.关键词
直肠癌/新辅助放化疗/机器学习/生物标志物/IFNGKey words
Rectal Cancer/Neoadjuvant Chemoradiotherapy/Machine Learning/Biomarkers/IFNG分类
医药卫生引用本文复制引用
杨玉玲,周瑛,吴宗妍,陈莎莎,李薇薇,符星..基于机器学习算法筛选直肠癌新辅助放化疗敏感相关基因[J].机器人外科学杂志（中英文）,2026,7(5):912-919,8.基金项目
安康职业技术学院2024年度科学研究项目(AZJKY2024004)2024 Scientific Research Projects of Ankang Vocational and Technical College(AZJKY2024004) （AZJKY2024004）
