单细胞测序结合机器学习算法揭示心力衰竭病人细胞间通讯及关键调控基因  

Single-cell Sequencing Combined with Machine Learning Algorithms Reveals Cell-cell Communication and Key Regulatory Genes in Patients with Heart Failure

作　　者：王学付 饶晋 张立 刘旭雯 张宇峰 WANG Xuefu;RAO Jin;ZHANG Li;LIU Xuwen;ZHANG Yufeng(School of Health Sciences and Engineering,University of Shanghai for Science and Technology,Shanghai 200093,China;Shanghai Changzheng Hospital,Shanghai 200003,China;School of Medicine,Guangxi University)

机构地区：[1]上海理工大学健康科学与工程学院,上海200093 [2]海军军医大学附属长征医院,上海200003 [3]广西大学医学院

出　　处：《中西医结合心脑血管病杂志》2025年第4期525-538,共14页Chinese Journal of Integrative Medicine on Cardio-Cerebrovascular Disease

基　　金：上海市曙光计划基金资助项目(No.21SG37)。

摘　　要：目的:基于单细胞测序和机器学习算法构建心力衰竭(HF)的诊断模型,并探索HF病人中细胞间通讯。方法:使用Seurat包对单细胞转录组测序(scRNA-seq)数据质控、降维、聚类和注释。通过AUCell评估各细胞亚群的免疫活性,选择免疫活性最高的细胞亚群进行后续分析。基于批量转录组测序(Bulk RNA-seq)数据,使用limma包筛选差异表达基因并进行基因集富集分析(GSEA)。进一步将疾病分类被作为反应变量,差异基因作为解释变量,通过4种机器学习模型来筛选具有诊断价值的巨噬细胞相关特征基因,并通过受试者工作特征(ROC)曲线评估关键基因的诊断能力。构建列线图预测HF发生的总风险分数。最后使用CellChat来探索细胞亚群之间的细胞间相互作用。结果:与正常样本相比,HF病人中巨噬细胞的比例高于正常样本,且巨噬细胞免疫活性评分最高。巨噬细胞亚群差异基因富集分析表明,白细胞介导的免疫过程和抗原的处理和呈递显著富集。多种机器学习算法相交结果发现SERPINA3、GPAT3、ANPEP和FCGBP可作为特征基因并与巨噬细胞密切相关。ROC曲线表明,诊断模型具有很好的预测能力。细胞通讯发现,由巨噬细胞移动抑制因子(MIF)介导的成纤维细胞-巨噬细胞以及ANNEXIN介导的巨噬细胞-中性粒细胞之间的信号通路表现出复杂的传出和传入动力学。结论:4个关键基因作为生物标志物具有良好的诊断价值。巨噬细胞介导的免疫过程以及细胞间通讯在HF的免疫微环境中起着关键作用。Objective:To construct a diagnostic model for heart failure(HF)based on single-cell sequencing and machine learning algorithms,and explore cell-cell communication in HF patients.Methods:The Seurat package was used for quality control,dimensionality reduction,clustering,and annotation of single-cell transcriptome(scRNA-seq)data.The immune activity of each cell subset was evaluated using AUCell,and the cell subset with the highest immune activity was selected for further analysis.Differential expression genes were screened using the limma package based on bulk transcriptome(Bulk RNA-seq)data,and gene set enrichment analysis(GSEA)was performed.Furthermore,disease classification was used as the response variable and differential genes as the explanatory variables to select macrophage-related feature genes with diagnostic value through four machine learning models.The diagnostic ability of key genes was evaluated using receiver operating characteristic(ROC)curves.Bar plots were also constructed to predict the overall risk score of HF occurrence.Finally,CellChat was used to explore cell-cell interactions between cell subtypes.Results:Compared to normal samples,the proportion of macrophages in HF patients was higher,and macrophages had the highest immune activity score.Gene enrichment analysis of macrophage subtypes showed significant enrichment of leukocyte-mediated immune processes and antigen processing and presentation.The intersection of multiple machine learning algorithms revealed that SERPINA3,GPAT3,ANPEP,and FCGBP could serve as feature genes and were closely related to macrophages.Receiver operator characteristic(ROC)curves demonstrated that our diagnostic model had good predictive ability.Cell communication analysis revealed complex outgoing and incoming dynamics in the signaling pathways between fibroblast-macrophage mediated-neutrophil mediated by MIF and macrophages--neutrophils mediated by ANNEXIN.Conclusion:The four key genes serve as biomarkers and have good diagnostic value.Macrophage-mediated immune p

关 键 词：心力衰竭 单细胞测序 机器学习 诊断模型 细胞通讯 

分 类 号：R73[医药卫生—肿瘤]