【摘要】 目的 鉴定阿尔茨海默病（AD）血-脑共性分子标志物并构建新型诊断模型，同时解析其与上游miRNA的潜在调控机制。方法 整合GEO平台16组公开数据（15组转录组：全血4组、脑组织11组；血清miRNA数据集1组：GSE120584）。采用SVA/ComBat进行批次校正，使用limma在全血与脑组织分别开展差异分析（|log₂FC|>0.2,FDR<0.05），并求交集定义候选共性分子标志物。以GSE63060+GSE63061为训练集，系统评估113种机器学习方法/组合构建诊断模型，在其余15个数据集进行外部验证。应用ClusterProfiler进行GO、KEGG富集，MANIA、Cytoscape构建PPI网络，CIBERSORT评估免疫浸润，miRDB、miRTarBase、TargetScan联合预测差异miRNA的高置信度靶基因。结果 经ComBat函数校正批次效应后对表达矩阵进行主成分分析，样本在PCA空间中显示分布更为一致；在全血数据中共鉴定DEGs 106个（上调20个，下调86个），在脑组织数据中共鉴定DEGs 2006个（上调934个，下调1072）（阈值：|log₂FC|>0.2,FDR<0.05），二者交集基因11个，差异方向一致基因8个，其中VCAN和FOS上调，NDUFA4、COX6C、HINT1、ACTR6、LSM3和ZC3H15下调，定义为候选血-脑共性分子标志物。GO富集分析显示，呼吸电子传递链、线粒体ATP合成耦联的电子传递及细胞色素c向氧的电子传递（复合体Ⅳ）显著富集（FDR<0.05）。KEGG分析显示，氧化磷酸化、产热与活性氧相关通路亦显著（FDR<0.05）；此外，AD、帕金森病、亨廷顿病、肌萎缩侧索硬化及朊蛋白病等神经变性疾病通路成簇富集；同时，非酒精性脂肪性肝病与糖尿病性心肌病亦被富集。以此8个基因构建的随机森林模型表现最佳：训练集AUC=0.999（95%CI:0.998～1.000），外部15个队列的AUC为0.932～1.000。PPI网络显示NDUFA4、COX6C为核心节点。CIBERSORT揭示AD外周免疫谱重塑（如中性粒细胞、M0巨噬细胞上升；幼稚B细胞、激活NK细胞下降）。miRNA预测交集提示miR-107的高置信度靶基因包含VCAN，指示“miR-107-VCAN-AD”潜在关键轴。结论 本研究在多队列、跨组织整合的框架下，预测到8个基因构成的血-脑共性分子标志物，并建立了在多外部队列稳定表现的诊断模型。机制证据一致指向线粒体OXPHOS/氧化应激与ECM/免疫通路，miR-107/VCAN轴为潜在上游调控靶点。该研究为AD的非侵入式分子诊断与机制研究提供了可推广的线索与工具。更多还原

【Abstract】 Objective To identify blood-brain common molecular markers for Alzheimer disease（AD） and construct a novel diagnostic model, while exploring the potential regulatory mechanism with upstream miRNAs. Methods Sixteen public datasets from the GEO platform were integrated（15 transcriptome datasets: 4 whole blood and 11 brain tissue; 1 serum miRNA dataset: GSE120584）. Batch correction was performed using SVA/ComBat,and differential expression analysis was conducted using limma in whole blood and brain tissue separately（|log₂ FC |>0.2, FDR <0.05）, with the intersection defining candidate common molecular markers. GSE63060+GSE63061 were used as the training set, and 113 machine learning methods/combinations were systematically evaluated to construct diagnostic models, which were externally validated in the remaining 15 datasets.ClusterProfiler was used for GO and KEGG enrichment analysis, MANIA and Cytoscape for PPI network construction, CIBERSORT for immune infiltration assessment, while miRDB and miRTarBase and TargetScan for joint prediction of high-confidence target genes of differentially expressed miRNAs. Results After the batch effect was corrected by the ComBat function, the expression matrix was subjected to principal component analysis, and the samples showed a more consistent distribution in the PCA space. A total of 106 DEGs（20 up-regulated and 86 down-regulated）were identified in whole blood data, and 2006 DEGs（934 up-regulated and 1072 down-regulated） were identified in brain tissue data（threshold: |log₂ FC|> 0.2, FDR<0.05）. There were 11 genes in the intersection of the two, and 8 genes in the same direction of difference. Among them, VCAN and FOS were up-regulated, and NDUFA4, COX6C, HINT1, ACTR6, LSM3 and ZC3H15 were down-regulated, which were defined as candidate bloodbrain common molecular markers. GO enrichment analysis showed that respiratory electron transport chain, electron transport coupled with mitochondrial ATP synthesis, and electron transport of cytochrome c to oxygen（complex Ⅳ） were significantly enriched（FDR<0.05）. KEGG analysis showed that oxidative phosphorylation, heat production and reactive oxygen species-related pathways were also significant（FDR<0.05）; in addition,neurodegenerative diseases such as AD, Parkinson′s disease, Huntington′s disease, amyotrophic lateral sclerosis and prion disease were clustered; at the same time, nonalcoholic fatty liver disease and diabetic cardiomyopathy were also enriched. The random forest model constructed with these 8 genes performed best: AUC=0.999（95%CI: 0.998-1.000） in the training set, and AUC was 0.932-1.000 in the external 15 cohorts. The PPI network showed NDUFA4 and COX6C as core nodes. CIBERSORT revealed peripheral immune profile remodeling in AD（e.g., increased neutrophils, M0 macrophages;decreased naive B cells, activated NK cells）. The intersection of miRNA predictions indicated that miR-107’s high-confidence target genes included VCAN, suggesting a potential key axis of "miR-107-VCAN-AD". Conclusion This study, within a multi-cohort and cross-tissue integration framework, identify 8 genes as blood-brain common molecular markers and establish a stable diagnostic model in multiple external cohorts.Mechanistic evidence consistently point to mitochondrial OXPHOS/oxidative stress and ECM/immune pathways, with the miR-107/VCAN axis as a potential upstream regulatory target. This study provides generalizable clues and tools for non-invasive molecular diagnosis and mechanism research of AD.更多还原