目的 通过整合网络推理算法、机器学习、类药性评估、分子对接和动力学模拟等方法，旨在筛选阿尔茨海默病（Alzheimer’s disease，AD）的生物标志物，阐明其发病机制，并探索其潜在治疗靶点及可入血成分（blood-absorbed constituents，BACs）。方法 基于中药入血成分及代谢产物数据库(Database of Constituents Absorbed Into Blood and Metabolites of Traditional Chinese Medicine，DCABM-TCM）筛选含有BACs的经典名方，通过证候本体与方剂数据库（Syndrome Ontology and Formula Database，SoFDA）检索经典名方所对应的证候及其靶点。利用基于加权有向图网络的推理（weighted signed directed tensor network-based inference，wSDTNBI）算法预测BACs靶点。通过基因表达综合数据库（gene expression omnibus，GEO）挖掘筛选AD差异性表达基因（differential gene expression，DEGs），并采用基因本体（gene ontology，GO）和京都基因与基因组百科全书（Kyoto encyclopedia of genes and genomes，KEGG）分析DEGs靶向证候基因所涉及参与的主要生物学过程和信号通路。采用最小绝对收缩和选择算法（least absolute shrinkage and selection operator，LASSO）、支持向量机递归特征消除（support vector machine-recursive feature elimination，SVM-RFE）方法、蛋白质-蛋白质相互作用网络（protein-protein interaction，PPI）和文本挖掘的方法筛选AD核心基因。结合泛分析干扰化合物（employ pan-assay interference compounds，PAINS）、Lipinski（Ro5）和Lipinski（Ro3）的过滤和分子对接来筛选候选BACs。结果 通过筛选193个经典名方，最终纳入10个含94个BACs的方剂及对应15种证候。预测得到1 520个证候基因和552个BACs靶点。进一步筛选出证候可靶向的528个上调及697个下调DEGs。富集分析显示DEGs主要参与神经元抗凋亡及突触功能等生物学过程，并显著富集于磷脂酰肌醇3-激酶（phosphatidylinositol 3-kinase，PI3K）-蛋白激酶B（protein kinase B，Akt）信号通路、黏着斑及AD发生通路。BACs-DEGs-AD网络表明上调和下调的DEGs分别可以靶向90和74个BACs，与9种证候相关。进一步通过PPI网络共分析得到度值较大的AD核心基因5个，分别是β2肾上腺素能受体（β2 adrenergic receptor，ADRB2）、P物质受体1（substance-P receptor 1，TACR1）、前列腺素G/H合酶2（prostaglandin G/H synthase 2，PTGS2）、丝氨酸蛋白酶HTRA1A（serine protease，HTRA1A）和代谢型谷氨酸受体1（metabotropic glutamate receptor 1，GRM1）。类药性评估筛选得到22个候选BACs，其中药理学文献验证有11个BACs具有抗AD活性。通过分子对接与动力学模拟结果表明，与上市药物多奈哌齐、加兰他敏和卡巴拉汀比较，unii-x87dcb9gst与5个AD核心基因中的乙酰胆碱脂酶（acetylcholinesterase，AChE）具有最稳定的综合结合能。结论 通过多模态算法筛选出AD的生物标志物，通过富集分析揭示AD相关的生物过程及信号通路，从分子层面阐释中医证候-AD基因的交互作用机制。同时，筛选得到的unii-x87dcb9gst可能为经典名方中治疗AD的候选BACs。不仅多维度解析AD发病的分子机制，更为抗AD药物研发提供创新性的生物标志物筛选体系和研究范式。

Objective This study aims to screen biomarkers for Alzheimer’s disease (AD), elucidate its pathogenesis, and explore potential therapeutic targets and blood-absorbed constituents (BACs) by integrating network inference algorithms, machine learning, drug-likeness evaluation, molecular docking, and molecular dynamics simulations. Methods Classical TCM formulas containing BACs were screened based on the database of constituents absorbed into the blood and metabolites of traditional Chinese medicine (DCABM-TCM) and the corresponding syndromes and their targets were retrieved from the from syndrome ontology to network-based evaluation of syndrome ontology and formula database (SoFDA). The weighted signed directed tensor network-based inference (wSDTNBI) algorithm was used to predict BACs targets. Differential expressed genes (DEGs) related to AD were identified by mining the GEO database. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) analyses were performed to explore the biological processes and pathways associated with up- and down-regulated DEGs. Core AD genes were screened using the least absolute shrinkage and selection operator (LASSO), support vector machine-recursive feature elimination (SVM-RFE), protein-protein interaction (PPI) network analysis, and text mining. Candidate BACs were further filtered using pan-assay interference compounds (PAINS), Lipinski’s Rule of Five (Ro5), and Rule of Three (Ro3), followed by molecular docking. Results First, a total of 10 classical TCM formulas containing 94 BACs and corresponding to 15 syndromes were selected from 193 formulas. A total of 1 520 syndrome-related genes and 552 BAC targets were predicted. Additionally, 528 up-regulated and 697 down-regulated DEGs targeted by syndromes were identified. Enrichment analysis revealed that these DEGs were primarily involved in biological processes such as positive regulation of gene expression, neuronal anti-apoptosis, and synaptic function, and were significantly enriched in pathways such as phosphatidylinositol 3-kinase (PI3K)-protein kinase B signaling pathway (Akt) signaling, focal adhesion, and AD pathways. The BACs-DEGs-AD network indicated that up- and down-regulated DEGs could target 90 and 74 BACs, respectively, associated with nine syndromes. PPI network analysis identified five core AD genes with high degrees: beta-2 adrenergic receptor (ADRB2), substance-P receptor (TACR1), prostaglandin G/H synthase 2 (PTGS2), serine protease HTRA1A, and metabotropic glutamate receptor 1 (GRM1). Drug-likeness evaluation screened 22 candidate BACs, 11 of which were pharmacologically validated to have anti-AD activity. Molecular docking results showed that unii-x87dcb9gst exhibited superior comprehensive binding energy with the five core AD genes compared to marketed drugs such as donepezil, galantamine, and rivastigmine. Finally, molecular dynamics simulations further confirmed the stable binding of unii-x87dcb9gst to the acetylcholinesterase (AChE) complex. Conclusion This study identified AD biomarkers through multimodal algorithms and revealed AD-related biological processes and signaling pathways through enrichment analysis, providing molecular insights into the interaction mechanisms between TCM syndromes and AD genes. Additionally, unii-x87dcb9gst, screened as a candidate BAC from classical TCM formulas, may serve as a potential therapeutic agent for AD. This research not only offers a multidimensional understanding of AD pathogenesis but also establishes an innovative biomarker screening system and research paradigm for anti-AD drug development.

TP18;R285

国家科技部重大新药创制项目（2017ZX09301001）；深圳市科技计划重点项目（JCYJ20220818101806014）；国家自然科学基金资助项目（81574038）；深圳大学第一附属医院有组织医学科学研究基金项目（2024YZZ11）