April 28, 2026 — Recently, a collaborative research team from MindRank AI, the University of Macau, and Imperial College London published a landmark study in the top international academic journal Nature Biomedical Engineering (Impact Factor 26.8), titled "DeepDrugDiscovery identifies blood–brain barrier permeable autophagy enhancers for Alzheimer’s disease".

The study successfully established an AI-powered platform for discovering active ingredients in Traditional Chinese Medicine (TCM) and natural products by integrating information from millions of compounds with multiple predictive modules. Utilizing this platform for virtual screening combined with cross-species experimental validation, the team identified small-molecule TCM compounds with therapeutic potential for Alzheimer’s disease (AD). This marks a milestone step in validating AI large models for empowering the development of new drugs for complex central nervous system (CNS) diseases.

Research Background: The "Death Valley" of AD Drug Development

Currently, there are over 50 million Alzheimer’s patients globally, yet therapeutic progress has remained slow. Dysfunctional autophagy, a key cellular cleaning process, is a primary pathological driver of AD. When this "scavenger system" of the brain is impaired, toxic proteins such as -amyloid (A) plaques and hyperphosphorylated tau proteins accumulate excessively. Consequently, developing autophagy enhancers is considered a highly forward-looking therapeutic strategy.

However, AD drug development faces immense technical barriers. Most existing autophagy enhancers rely on the mTOR-dependent pathway, which acts as a central hub for human metabolism and easily triggers off-target side effects that interfere with normal physiological homeostasis. Furthermore, the obstruction of the blood–brain barrier (BBB), the complexity of AD pathogenesis, and the lack of effective autophagic biomarkers in the brain collectively constitute the "Death Valley" of AD drug discovery.

DeepDrugDiscovery: Leading a Paradigm Shift in Screening

To tackle this global medical challenge, the research team introduced cutting-edge AI technology. Led by MindRank AI, the DeepDrugDiscovery platform was constructed to perform an "underlying reconstruction" of the drug discovery process for brain diseases.

The platform utilizes a hybrid molecular representation architecture integrating a variational autoencoder (VAE) and a gated recurrent unit (GRU). It encodes 2,048-bit Morgan fingerprints (ECFP4) alongside 19 one-dimensional (1D) and three-dimensional (3D) descriptors, learning informative molecular latent representations through unsupervised pre-training.

Unlike traditional screening that relies on a single target or structural similarity, this technology focuses on a "mechanism-aware" (or mechanism-centric) approach, achieving a paradigm-level elevation in screening precision and efficiency.

The team first applied DeepDrugDiscovery to screen 1.16 million compounds from the University of Macau's vast library of natural products and TCM compounds. Utilizing a GPU-accelerated molecular attention mechanism, the platform completed a massive similarity matrix calculation of 50 × 1,155,606 in an extremely short time, identifying 6,834 initial hits.

Subsequently, MindRank’s ADMET Ranker™ (an advanced graph Transformer module) was used to simultaneously evaluate critical drug-like properties, including BBB permeability, Caco-2, MDCK, LogD, pKa, and solubility. The team successfully narrowed the candidates to 449 high-potential molecules. Combined with molecular docking against targets like FKBP12, mTOR kinase, and the FKBP12–mTOR complex, as well as verification of commercial availability, 15 top candidates were finally selected for experimental validation.

AI Empowerment: Efficient and Precise Identification of Lead Compounds

Experimental validation confirmed that all 15 candidates demonstrated the ability to promote autophagy in cell assays. The team further analyzed changes in autophagic markers in N2a cells and monitored mTOR signaling, identifying 7 mTOR-independent candidates that enhance autophagy without significantly affecting the mTOR pathway.

Based on novelty in mechanism, innovative neuroprotective activity, and chemical structural diversity, the team selected four candidates for in-depth study. In AD-related cell models, all four molecules promoted the clearance of abnormal proteins. Among them, Ombuin (Omb) and 2-Hydroxycinnamic acid (2-HCA) performed most prominently and were identified as core lead compounds.

These two leads underwent in vivo validation in C. elegans and 3×Tg-AD mouse models, and further assessment of their blood–brain barrier penetration confirmed their strong neuroprotective potential.

Significance and Outlook

The corresponding authors of this study are Professor Jia-Hong Lu of the University of Macau and Dr. Zhangming Niu, founder and CEO of MindRank AI. The co-first authors include Yu Dong, Xu-Xu Zhuang, Xianglu Xiao, and Wenfan Wu. Significant support was also provided by Han-Ming Shen, Jian-Bo Wan, Huanxing Su, Hua Yu, and Defang Ouyang from the University of Macau.

The research team has open-sourced the platform (https://deepdrugdiscovery.mindrank.ai) to empower further TCM mechanism analysis and innovative drug development.

This study not only contributes two highly promising candidate molecules for the treatment of Alzheimer’s disease but, more importantly, validates a highly viable, reproducible, and scalable AI-driven drug discovery path. By efficiently integrating mechanism-oriented screening, ADMET evaluation, and cross-species experimental validation, AI large model technology can significantly shorten the early discovery cycle, reduce costs, and substantially improve the success rate in high-difficulty fields such as the central nervous system.