Ancient Chinese Formula Decoded: Machine Learning Reveals How Bushen-Yizhi May Fight Alzheimer’s

Alzheimer’s disease remains one of medicine’s most stubborn adversaries, an incurable neurodegenerative condition that steadily strips away memory and cognition despite decades of drug development. Now a team of Chinese researchers has taken an unusually modern approach to an ancient remedy, combining high-resolution chemistry, machine learning, network biology and laboratory experiments to work out exactly how a traditional herbal formula called Bushen-Yizhi, or BSYZ, might protect the brain. The study, published in BMC Complementary Medicine and Therapies, offers one of the most complete mechanistic portraits yet of a multi-herb formula acting on Alzheimer’s-like cognitive impairment, and it demonstrates how computational pipelines can transform folk medicine into testable, target-level science.

The researchers began with chemistry. Using ultra-performance liquid chromatography coupled with quadrupole time-of-flight tandem mass spectrometry, a technique capable of separating and identifying molecules with exquisite precision, they characterized the chemical composition of BSYZ. The analysis identified 105 distinct compounds in the formula, and supplementary data showed which of these were absorbed into the blood and which crossed into the brain, a crucial distinction for any candidate therapy targeting the central nervous system. The 105 compounds were distributed across the six herbs that make up the formula, giving the team a complete molecular inventory of what the remedy actually delivers to the body rather than relying on assumptions drawn from historical texts.

With the molecular inventory in hand, the team turned to biology. They mined publicly available transcriptomic datasets from the Gene Expression Omnibus, performing differential expression analysis and weighted gene co-expression network analysis, a statistical method that identifies groups of genes whose activity rises and falls together across disease states. This bioinformatic sweep uncovered 81 potential Alzheimer’s-related targets that the formula’s compounds might plausibly engage. But 81 targets is too many for a meaningful mechanistic story, so the researchers deployed an unusually rigorous filtering strategy: they screened the candidate list using 127 different machine learning models, then interpreted the consensus results with SHAP analysis, a technique borrowed from explainable artificial intelligence that quantifies how much each feature contributes to a model’s predictions.

The machine learning gauntlet distilled the field down to seven core genes: HIPK2, CXCR4, CCKBR, PTPN13, SFN, ALB and ALDH1A1. Each of these plays a documented role in Alzheimer’s-related pathological processes, from neuroinflammation and oxidative stress to synaptic dysfunction. The team then validated the computational predictions experimentally using RT-qPCR, a laboratory technique that measures messenger RNA levels and confirms whether the predicted genes are genuinely differentially expressed in disease. They were. Moreover, a stage-stratified transcriptomic analysis revealed that the expression of these seven genes changes dynamically as Alzheimer’s disease progresses, suggesting the formula’s targets are not static markers but active participants in the disease trajectory.

The biological effects of the formula were tested in a well-established animal model. Mice were treated with scopolamine, a drug that blocks acetylcholine signaling and reliably induces Alzheimer’s-like cognitive impairment, providing a fast and reproducible way to study memory deficits. When scopolamine-treated mice received BSYZ, the results were striking across multiple measures. In the Morris water maze, a standard test in which rodents must learn the location of a hidden platform, the treated animals showed ameliorated cognitive deficits compared with untreated scopolamine controls. Open field testing and histopathological examination using haematoxylin-eosin staining added further behavioral and structural evidence of neuroprotection.

Biochemical assays filled in the molecular picture behind those behavioral improvements. The researchers measured markers of oxidative stress, including superoxide dismutase and glutathione peroxidase, two antioxidant enzymes that protect neurons from damaging reactive molecules, and malondialdehyde, a byproduct of lipid peroxidation that serves as a fingerprint of oxidative damage. BSYZ treatment improved the antioxidant profile, indicating reduced oxidative stress. The formula also corrected cholinergic dysfunction, the loss of acetylcholine signaling that scopolamine induces and that mirrors a hallmark of Alzheimer’s pathology, by modulating acetylcholine, acetylcholinesterase and choline acetyltransferase. Finally, the treatment dampened neuroinflammation, the chronic immune activation in the brain that is increasingly recognized as a central driver of neurodegeneration.

To connect the seven core genes back to specific chemical compounds, the team employed molecular docking, a computational method that predicts how small molecules fit into the binding pockets of proteins, followed by 100-nanosecond molecular dynamics simulations that test whether those predicted complexes remain stable over time. The simulations tracked metrics such as root mean square deviation, root mean square fluctuations, radius of gyration and solvent-accessible surface area, all standard indicators of whether a protein-ligand complex is structurally stable or falling apart. The computational evidence pointed to two compounds in particular: isoquercitrin, a flavonoid glycoside, and epicatechin gallate, a polyphenol better known as a major active constituent of green tea. Both appeared to bind stably to CXCR4, a chemokine receptor, and CCKBR, the cholecystokinin B receptor, providing a plausible chemical-to-target link for two of the seven core genes.

The significance of this work extends well beyond a single herbal formula. Traditional Chinese medicine has long been criticized in Western pharmacology for its complexity, since multi-herb formulas contain hundreds of compounds acting on dozens of targets simultaneously, defying the one-drug-one-target paradigm that dominates pharmaceutical development. This study shows how that complexity can be tamed rather than dismissed. By combining untargeted chemical profiling with network analysis, machine learning consensus screening, explainable AI interpretation, transcriptomic validation and molecular simulation, the researchers built a framework in which every step is auditable and every conclusion is anchored to experimental data. The result is a mechanistic hypothesis precise enough to guide preclinical development.

The authors are careful about scope, and so should readers be. The animal model used here, scopolamine-induced cognitive impairment, captures certain features of Alzheimer’s disease but does not reproduce the full pathology of amyloid plaques and tau tangles that define the human condition. The molecular docking and dynamics results are computational evidence, not proof of binding in living tissue, and the RT-qPCR validation confirms gene expression changes rather than direct compound-target engagement in the brain. The paper itself was shared early as an accepted manuscript subject to further edits, carrying a permanent DOI and citable status. These caveats are standard for the field, but they mean the findings should be read as a rigorous foundation for further investigation rather than a clinical endorsement.

Nevertheless, the trajectory the study maps out is clear and potentially impactful. The demonstration that BSYZ acts through a multi-component, multi-target mechanism, engaging seven core genes that shift dynamically across disease stages, aligns with the growing recognition that complex neurodegenerative diseases may require correspondingly complex therapeutic interventions. The identification of isoquercitrin and epicatechin gallate as candidate bioactive molecules gives medicinal chemists concrete starting points for isolation, synthesis and optimization. And the systems pharmacology framework itself, integrating mass spectrometry, machine learning and experimental validation, is a reusable template that other research groups can apply to any traditional formula. For a disease that has defeated hundreds of single-target drug candidates, the message from this study is that ancient polypharmacy, viewed through the lens of modern computational biology, may hold lessons that reductionist approaches have missed.

Subject of Research: Systems pharmacology investigation of the traditional Chinese medicine formula Bushen-Yizhi as a multi-target therapy for Alzheimer's disease

Article Title: A system pharmacology framework to explore the therapeutic mechanism of Bushen-Yizhi formula against Alzheimer’s disease via integrating UPLC-Q-TOF-MS/MS, network analysis, machine learning and experimental validation

Article References: Zhang, J., Rao, H., Zhang, X., Zhao, S., Dai, Z., Tang, X., Cai, C., An, Y., Fang, S., Zhuo, Y., Li, H., & Fang, J. (2026). A system pharmacology framework to explore the therapeutic mechanism of Bushen-Yizhi formula against Alzheimer’s disease via integrating UPLC-Q-TOF-MS/MS, network analysis, machine learning and experimental validation. BMC Complementary Medicine and Therapies. https://doi.org/10.1186/s12906-026-05632-8

Image Credits: AI Generated

DOI: 10.1186/s12906-026-05632-8

Keywords: Alzheimer's disease, Bushen-Yizhi formula, systems pharmacology, machine learning, UPLC-Q-TOF-MS/MS, molecular docking, molecular dynamics simulation, traditional Chinese medicine, neuroinflammation, oxidative stress, network analysis, scopolamine mouse model