目的 通过整合多组学与机器学习方法，筛选缺血性卒中（ischemic stroke，IS）关键基因并解析其机制，预测其潜在的中医药防治靶点。方法 整合GEO数据库中IS转录组数据，经差异表达、加权基因共表达网络分析（weighted gene co-expression network analysis，WGCNA）及蛋白互作网络分析筛选候选基因。运用包括最小绝对收缩和选择算子（logistic least absolute shrinkage and selection operator，LASSO）、随机森林等在内的多种机器学习算法构建诊断模型，并通过交叉验证确定最优基因集。通过构建大鼠大脑中动脉栓塞（middle cerebral artery occlusion，MCAO）模型，以贝德森（Bederson）评分、HE染色及实时荧光定量聚合酶链反应（real-time quantitative polymerase chain reaction，RT-qPCR）进一步验证。利用CIBERSORTx分析免疫浸润，并基于Coremine Medical数据库反向匹配潜在中药。结果 确定8个核心基因ARG1、CLEC4E、CLEC5A、FCAR、FCGR1A、IRAK3、MCEMP1、TLR5，其诊断模型在训练与验证队列中均表现良好［曲线下面积（area under curve，AUC）＞0.7］。动物实验证实上述基因在IS模型大鼠皮层组织中表达显著上调，且与M0型巨噬细胞、中性粒细胞等免疫细胞浸润水平密切相关。基于上述靶点预测出59味潜在中药，四气多属寒、温、平，五味多属苦、甘，归经主要集中在肝、肾和脾经，其性味归经与IS“肝肾阴虚、瘀热内蕴”病机相符。结论 通过整合生物信息学、机器学习与实验验证，系统筛选并验证了8个参与卒中后免疫微环境重塑的关键基因，可作为IS的潜在诊断生物标志物。中药预测分析提示靶向这些基因的中药性味归经特点与IS病机相符，为从免疫调控角度开展IS的中医药防治提供了理论依据。

Objective To screen key genes of ischemic stroke (IS) and analyze their mechanisms by integrating multi-omics and machine learning methods, and predict their potential targets for traditional Chinese medicine prevention and treatment. Methods IS transcriptome data from GEO database was integrated, and candidate genes were screened through differential expression, weighted gene co-expression network analysis (WGCNA) and protein interaction network analysis. A variety of machine learning algorithms including logistic least absolute shrinkage and selection operator (Lasso), random forest, etc. were used to construct a diagnostic model, and the optimal gene set is determined through cross-verification. An middle cerebral artery occlusion (MCAO) model was constructed and further verified by Bederson scoring, HE staining and real-time quantitative polymerase chain reaction (RT-qPCR). Immune infiltration was analyzed using CIBERSORTx and potential traditional Chinese medicines were reverse-matched based on Coremine Medical database. Results Eight core genes (ARG1, CLEC4E, CLEC5A, FCAR, FCGR1A, IRAK3, MCEMP1, TLR5) were identified, and their diagnostic models performed well in both the training and verification cohorts [area under curve (AUC) > 0.7]. Animal experiments had confirmed that the expression of these genes was significantly up-regulated in the cortical tissue of IS model rats, and was closely related to the infiltration level of immune cells such as M0-type macrophages and neutrophils. Based on the above targets, 59 potential traditional Chinese medicines were predicted. Most of the four qi were cold, warm and calm, and most of the five are bitter and sweet. The meridian tropism was mainly concentrated in the liver, kidney and spleen meridians. The medicinal properties and meridian tropism were consistent with the pathogenesis of IS “liver and kidney yin deficiency, and stasis and heat accumulation”. Conclusion By integrating bioinformatics, machine learning and experimental validation, this study systematically identified and validated eight key genes involved in the remodelling of the post-stroke immune microenvironment, which may serve as potential diagnostic biomarkers for IS. Predictive analysis of traditional Chinese medicine suggests that the pharmacological properties—including taste, nature and meridian tropism—of drugs targeting these genes are consistent with the pathogenesis of IS, thereby providing a theoretical basis for the prevention and treatment of IS using traditional Chinese medicine from the perspective of immune regulation.

Q811.4;R285

国家自然科学项目基金(82560989); 柳州市科技项目(2024YB0103B011); 广西青年岐黄学者培养计划(GXQH202414); 八桂青年拔尖人才项目(桂人才办[2025]1号)