人工智能
中医药
机器学习
疾病
计算机科学
计算生物学
生物信息学
医学
精密医学
人血浆
人工神经网络
深度学习
作者
Heng Lin,Zixiong Huang,Zhaungbin Liao,Qibiao Wu,Lili Cui
标识
DOI:10.1016/j.brainresbull.2026.112221
摘要
This study integrated multi-omics data and machine learning to prioritize candidate plasma proteins and to predict the multi-target interactions of traditional Chinese medicine (TCM) in Alzheimer's disease (AD). Mendelian randomization across three cohorts (Icelandic, UK Biobank, and Fenland) identified 66 AD-associated plasma proteins, including TREM2, ACE, GRN, and CR1, with TREM2 showing a robust protective effect (OR = 0.78, 95% CI: 0.75-0.81). Single-cell RNA sequencing showed neuron-specific downregulation of NSF and TMEM106B in AD, implicating synaptic dysfunction and impaired axonal transport. A machine-learning classifier (K-nearest neighbors, KNN) incorporating TREM2, TMEM106B, C1S, PLOD2, and NSF, trained on brain-tissue expression data, demonstrated moderate predictive performance (AUC = 0.76, 95% CI: 0.63-0.90) and is presented as a candidate brain-tissue-derived classifier rather than a validated non-invasive plasma diagnostic panel. Immune infiltration analysis revealed altered M2 macrophage and plasma cell dynamics, with NSF negatively correlated with γδ T-cell infiltration (p < 0.001). Network pharmacology prioritized five candidate TCM herbs (e.g., Scutellaria baicalensis [Huang Qin] and Alpinia oxyphylla [Yi Zhi]) and bioactive compounds (stigmasterol and quercetin) targeting AD-related proteins. Molecular docking predicted strong binding between stigmasterol and PLOD2 (binding energy: -11.2kcal/mol), suggesting a potential role in modulating extracellular matrix remodeling. These compounds were enriched in the PI3K-Akt and NF-κB pathways, consistent with the anti-inflammatory effects observed in TCM formulations such as GHYSJ. These findings link candidate biomarker discovery with the predicted multi-target interactions of TCM, providing hypotheses for future non-invasive diagnosis and precision therapy in AD.
科研通智能强力驱动
Strongly Powered by AbleSci AI