Data Sheet 5_Integration of multi-omics and machine learning strategies identifies immune related candidate biomarkers in inflammation-associated hypertrophic cardiomyopathy.csv

BackgroundHypertrophic cardiomyopathy (HCM) is a common inherited heart disease frequently leading to heart failure. Although sarcomeric gene mutations are known, they only account for a subset of cases. The role of immune dysregulation in HCM progression has gained increasing attention, necessitating the exploration of immune-related biomarkers and therapeutic targets. This study integrates Mendelian randomization (MR), transcriptomics, machine learning, and experimental validation to investigate the immune mechanisms underlying HCM.
MethodsWe analyzed three transcriptomic datasets from the GEO database (210 healthy controls, 152 HCM patients) and identified differentially expressed genes (DEGs) using the R package limma. MR analysis was performed on 19,942 expression quantitative trait loci (eQTLs) and HCM cases using the TwoSampleMR package. Machine learning (10 algorithms) was employed to construct diagnostic models, and SHAP analysis was applied to assess key gene contributions. Functional enrichment was performed with clusterProfiler, diagnostic performance was evaluated via ROC curves, and immune cell infiltration was analyzed using CIBERSORT. A competing endogenous RNA (ceRNA) network was constructed, and drug targets were predicted via the DGIdb database. Key gene expression was validated by qPCR.
ResultsWe identified 472 DEGs and 205 HCM-associated loci, narrowing down to seven key genes: RNF165, SNCA, SRGN, MARCO, STEAP4, SIGLEC9, and TKT. These genes were enriched in immune-related pathways (e.g., cytokine activity, leukocyte migration, JAK-STAT signaling). The Random Forest model exhibited the highest diagnostic performance (AUC: 0.939), with SHAP analysis revealing MARCO as the top contributor. Gene expression was associated with immune cell infiltration: HCM samples showed increased CD4+ T cells and M0 macrophages but decreased M2 macrophages and neutrophils. The ceRNA network comprised 5 mRNAs, 40 miRNAs, and 152 lncRNAs. SRGN and SNCA were identified as potential targets for heparin and 33 other drugs, respectively. qRT-PCR performed on a small number of myocardial samples supported expression trends of the identified genes, in line with transcriptomic analysis.
ConclusionThis study reveals immune-related mechanistic biomarkers and potential therapeutic targets in HCM, highlighting the role of immune dysregulation in disease progression. Machine learning and SHAP analysis improved diagnostic model interpretability, providing a basis for future development of non-invasive diagnostic tools.

背景 肥厚型心肌病（Hypertrophic cardiomyopathy, HCM）是一种常见的遗传性心脏病，常可进展为心力衰竭。尽管目前已明确肌节基因突变与该病相关，但仅能解释部分病例。免疫失调在HCM病程中的作用日益受到关注，因此亟需探索免疫相关生物标志物与治疗靶点。本研究整合孟德尔随机化（Mendelian randomization, MR）、转录组学、机器学习及实验验证手段，探究HCM潜在的免疫机制。
方法 我们分析了来自GEO数据库的3组转录组数据集（包含210名健康对照者、152名HCM患者），并通过R包limma筛选差异表达基因（differentially expressed genes, DEGs）。采用TwoSampleMR包对19942个表达数量性状位点（expression quantitative trait loci, eQTLs）与HCM病例开展孟德尔随机化分析。借助10种机器学习算法构建诊断模型，并通过SHAP分析评估关键基因的贡献度。运用clusterProfiler进行功能富集分析，通过ROC曲线评估诊断性能，采用CIBERSORT分析免疫细胞浸润情况。构建内源竞争RNA（competing endogenous RNA, ceRNA）调控网络，并通过DGIdb数据库预测药物靶点。最后通过定量聚合酶链反应（quantitative PCR, qPCR）验证关键基因的表达水平。
结果 本研究共筛选得到472个差异表达基因与205个HCM相关位点，最终锁定7个关键基因：RNF165、SNCA、SRGN、MARCO、STEAP4、SIGLEC9及TKT。这些基因显著富集于免疫相关通路（如细胞因子活性、白细胞迁移、JAK-STAT信号通路）。随机森林模型展现出最优的诊断性能（AUC：0.939），SHAP分析显示MARCO为贡献度最高的基因。基因表达水平与免疫细胞浸润状态显著相关：HCM样本中CD4+ T细胞与M0巨噬细胞浸润增加，而M2巨噬细胞与中性粒细胞浸润减少。本研究构建的ceRNA网络包含5个mRNA、40个miRNA及152个lncRNA。SRGN与SNCA分别被预测为肝素及其他33种药物的潜在靶点。对少量心肌样本开展逆转录定量聚合酶链反应（quantitative reverse transcription PCR, qRT-PCR）验证的结果显示，关键基因的表达趋势与转录组分析结果一致。
结论 本研究揭示了HCM中与免疫相关的机制性生物标志物与潜在治疗靶点，阐明了免疫失调在疾病进展中的作用。机器学习与SHAP分析提升了诊断模型的可解释性，为未来开发非侵入性诊断工具提供了理论依据。