机器学习挖掘环境污染物介导的胶质母细胞瘤核心基因：诊断、复发与预后生物标志物
Machine learning models decipher environmental pollutant-driven core genes in glioblastoma: Biomarkers for diagnosis, recurrence, and prognosis
Junjia Pan∗∗, Xuanyao Yuan∗∗, Haoshen Lin∗∗, Rui Zhao, Lin Yang∗, Chun Hu∗
https://doi.org/10.1016/j.jes.2025.11.028
摘要
工业生产、汽车尾气排放带来的各类环境污染，现已被证实是诱发胶质母细胞瘤（GBM）等多种疾病的重要危险因素。但目前尚未有研究系统解析环境污染物暴露与胶质母细胞瘤发病的关联。本研究整合环境暴露数据与临床样本数据，借助多种机器学习算法筛选核心基因，并构建三类分别用于胶质母细胞瘤诊断、复发预测及预后评估的预测模型。研究采用随机森林、LASSO 回归、逻辑回归与 Cox 回归模型完成建模，并通过独立外部数据集验证模型效能；同时开展 GO、KEGG 功能富集分析，探究潜在分子调控机制；利用分子对接技术验证关键基因与大气污染物的相互作用。研究结果显示，多环芳烃、氮氧化物及可吸入颗粒物可通过调控细胞增殖相关通路，促进胶质母细胞瘤的发生与进展。结合机器学习筛选与分子对接验证，本研究筛选出多组关键基因：用于发病预测的NFKBIA、BCL2L12；用于复发预测的DKK3、FGFR1、GLIPR1、TRIM8；以及用于预后评估的STAT3、NF1、KDM5A。上述基因有望成为胶质母细胞瘤早期诊断、复发监测及预后判断的潜在生物标志物。本研究证实了环境 - 基因互作在胶质母细胞瘤病程中的关键作用，为该病临床诊断与治疗方案研发提供了理论依据。
亮点
本研究依托机器学习与分子对接解析大气污染物促胶质母细胞瘤进展的分子通路，筛选出适配发病预测、复发监测、预后评估的分组关键基因，证实环境 - 基因互作的致病价值，为胶质母细胞瘤临床诊疗提供新型标志物与理论依据。
Abstract
Environmental pollution, particularly from industrial production and automobile exhaust emissions, is increasingly recognized as a significant factor in the development of various diseases, including glioblastoma (GBM). However, the relationship between environmental exposure and GBM has not been systematically analyzed. In this study, we utilized machine learning techniques to develop three predictive models for GBM diagnosis, recurrence, and prognosis, considering core genes identified through environmental and clinical data. We applied Random Forest, LASSO regression, logistic regression, and Cox regression models, and validated them with independent external datasets. GO and KEGG functional enrichment analyses were also performed to explore the potential underlying mechanisms. Molecular docking was used to examine interactions between key genes and air pollutants. Our results indicated that polycyclic aromatic hydrocarbons, nitrogen oxides, and particulate matter contribute to the onset and progression of GBM by affecting cell proliferation pathways. Using machine learning and molecular docking, we identified several key genes: NFKBIA and BCL2L12 for onset prediction; DKK3, FGFR1, GLIPR1, and TRIM8 for recurrence prediction; and STAT3, NF1, and KDM5A for prognosis prediction. These genes may serve as potential biomarkers for early diagnosis, recurrence monitoring, and prognosis of GBM. This study underscores the importance of environmental gene interactions in GBM, offering valuable insights into clinical diagnosis and treatment strategies.
作者简介
第一作者：潘俊佳硕士，2026年于华南师范大学脑科学与康复医学研究院获得硕士学位。主要从事生物计算科学与神经系统疾病的相关研究。在Journal of Environmental Sciences、International Journal of Biological Macromolecules、iScience和Molecular Neurobiology等期刊发表多篇研究论文。
通讯作者：胡春，华南师范大学脑科学与康复医学研究院教授，博士研究生导师。广东省脑认知与人的素质发展基础学科研究中心成员。2016年于德国汉堡大学获哲学博士学位。主要从事神经发育与疾病的相关研究。在Nature Neuroscience、J Neuroscience、Neuroscience Bulletin、International Journal of Biological Macromolecules和Molecular Neurobiology等期刊发表学术论文。
文章来源：https://url.scnu.edu.cn/record/view/index.html?key=47e9cbe66d6cfb2f9984f45c90191e13