目的 利用生物信息学及机器学习方法探究黄芪甲苷治疗扩张型心肌病（DCM）的分子机制。方法 从 高通量基因表达数据库（GEO）获取 DCM 转录组数据集（GSE9800、GSE29819），筛选显著差异表达基因 （DEGs）。基于 ChEMBL、SwissTargetPrediction、PharmMapper 和 Comparative Toxicogenomics Database 数据库系 统筛选黄芪甲苷的作用靶点，对 DEGs 与黄芪甲苷的预测靶点取交集，交集靶点即为黄芪甲苷治疗 DCM 的潜 在作用靶点。对潜在作用靶点通过 GO 功能注释与 KEGG 通路富集分析其涉及的生物学过程和信号通路。构建 127 种机器学习模型识别特征基因，并应用沙普利值加法解释算法（SHAP）评价靶点特征重要性。利用受试者 工作特征曲线（ROC）评估特征基因的疾病状态预测效能，并进行外部验证。使用分子对接技术分析黄芪甲苷与 特征基因的结合活性，并对高亲和力基因进一步实施孟德尔随机化（MR）因果推断验证。结果 共筛选出黄芪 甲苷潜在作用靶点 296 个，DCM 的 DEGs 959 个，得到 25 个交集靶点。GO 分析显示，交集靶点主要参与氨 酸型肽酶活性、丝氨酸水解酶活性、血压调节、激素代谢、血小板活性等生物学功能。KEGG 分析显示，交集 靶点主要参与细胞色素 P450 相关通路、钙信号通路、酪氨酸代谢、脂肪酸降解、丙酮酸代谢、糖酵解/糖异 生、肾素-血管紧张素系统（RAS）、冠状病毒病（COVID-19）、腺苷酸活化蛋白激酶（AMPK）信号通路、环磷酸 鸟苷-蛋白激酶 G（cGMP-PKG）信号通路等。基于机器学习算法进一步筛选出 10 个特征基因（BMP7、NPR3、 ADRB1、SLC37A4、FAP、CA12、DPP9、CRAT、PTPN22 及 DAPK1）。ROC 分析证实，所有特征基因对 DCM 均有良好的预测能力，且外部验证具有显著组间差异（P＜0.05）。分子对接显示，肉碱乙酰转移酶（CRAT）、死 亡相关蛋白激酶 1（DAPK1）、β1 肾上腺素能受体（ADRB1）与黄芪甲苷的结合能均＜-10 kcal·mol-1 ，表明存在 高亲和力结合。MR 分析显示，CRAT、DAPK1 与 DCM 存在显著因果关联（P＜0.05）。结论 黄芪甲苷可能通 过调控代谢重编程、维持钙稳态及干预神经内分泌通路等多途径发挥对 DCM 的心脏保护作用，其中 CRAT、 DAPK1 为其关键作用靶点，可为阐明 DCM 发病机制及创新药物研发提供分子依据。

Objective To investigate the molecular mechanisms of astragaloside IV in the treatment of dilated cardiomyopathy （DCM） using bioinformatics and machine learning approaches. Methods DCM transcriptome datasets （GSE9800， GSE29819） were obtained from the Gene Expression Omnibus （GEO） database， and significantly differentially expressed genes （DEGs） were identified. Putative targets of astragaloside IV were systematically screened using the ChEMBL，SwissTargetPrediction，PharmMapper，and Comparative Toxicogenomics Database. Overlapping targets between the DEGs and the predicted astragaloside IV targets were identified as potential therapeutic targets of astragaloside IV for DCM. Gene Ontology （GO） functional annotation and Kyoto Encyclopedia of Genes and Genomes （KEGG） pathway enrichment analyses were performed to explore the biological processes and signaling pathways associated with these potential targets. A panel of 127 machine learning models was constructed to identify feature genes， and the SHapley Additive exPlanations （SHAP） algorithm was applied to evaluate the importance of these target features. The predictive performance of the feature genes for disease status was assessed using receiver operating characteristic （ROC） curves，followed by external validation. Molecular docking was employed to analyze the binding affinity between astragaloside IV and the feature genes， and genes exhibiting high affinity were further subjected to causal inference validation using Mendelian randomization （MR）. Results A total of 296 potential therapeutic targets of astragaloside IV and 959 DEGs for DCM were identified，yielding 25 overlapping targets. GO analysis indicated that the overlapping targets were primarily involved in biological functions such as serine-type peptidase activity， serine hydrolase activity，blood pressure regulation，hormone metabolism，and platelet activity. KEGG analysis revealed that the overlapping targets were mainly enriched in pathways including cytochrome P450-related pathways， calcium signaling pathway， tyrosine metabolism， fatty acid degradation， pyruvate metabolism， glycolysis/gluconeogenesis， renin-angiotensin system （RAS）， coronavirus disease （COVID-19）， AMP-activated protein kinase （AMPK） signaling pathway， and cyclic guanosine monophosphate-protein kinase G （cGMP-PKG） signaling pathway. Ten feature genes （BMP7，NPR3，ADRB1，SLC37A4，FAP，CA12，DPP9，CRAT，PTPN22，and DAPK1） were further identified based on machine learning algorithms. ROC analysis confirmed that all feature genes possessed good predictive ability for DCM， and external validation showed significant inter-group differences （P＜0.05）. Molecular docking revealed that the binding energies of carnitine acetyltransferase （CRAT），death-associated protein kinase 1 （DAPK1），and β1 adrenergic receptor （ADRB1） with astragaloside IV were all less than -10 kcal·mol-1 ，indicating high-affinity binding. MR analysis demonstrated significant causal associations between CRAT and DAPK1 with DCM （P＜0.05）. Conclusion Astragaloside IV may exert cardioprotective effects in DCM through multiple pathways， including regulating metabolic reprogramming， maintaining calcium homeostasis， and modulating neuroendocrine pathways. CRAT and DAPK1 are identified as key targets of its action，providing a molecular basis for elucidating the pathogenesis of DCM and for the development of innovative therapies.

R285.5；R857.3

国家自然科学基金项目（81703894）；北京市自然科学基金青年科学基金项目（7254523）。