2. Academic Validation
  3. AI-identified CD133-targeting natural compounds demonstrate differential anti-tumor effects and mechanisms in pan-cancer models

AI-identified CD133-targeting natural compounds demonstrate differential anti-tumor effects and mechanisms in pan-cancer models

  • EMBO Mol Med. 2025 Oct 2. doi: 10.1038/s44321-025-00308-1.
Yibo Hou # 1,  Zixian Wang # 1,  Wenlin Wang # 2,  Qing Tang 2,  Yongde Cai 3,  Siyang Yu 1,  Jin Wang 3,  Xiu Yan 3,  Guocai Wang 2,  Peter E Lobie 1,  Yubo Zhang 4,  Xiaoyong Dai 5,  Shaohua Ma 6
Affiliations

Affiliations

  • 1 Institute of Biopharmaceutical and Health Engineering, Tsinghua Shenzhen International Graduate School (SIGS), Tsinghua University, Shenzhen, 518055, China.
  • 2 Department of Physiology, School of Medicine; Institute of Traditional Chinese Medicine & Natural Products, College of Pharmacy, and Guangdong Province Key Laboratory of Pharmacodynamic Constituents of TCM and New Drugs Research, Jinan University, Guangzhou, 510632, China.
  • 3 Synorg Biotechnology (Shenzhen) Co. Ltd., Shenzhen, 518107, China.
  • 4 Department of Physiology, School of Medicine; Institute of Traditional Chinese Medicine & Natural Products, College of Pharmacy, and Guangdong Province Key Laboratory of Pharmacodynamic Constituents of TCM and New Drugs Research, Jinan University, Guangzhou, 510632, China. ybzhang99@jnu.edu.cn.
  • 5 Department of Physiology, School of Medicine; Institute of Traditional Chinese Medicine & Natural Products, College of Pharmacy, and Guangdong Province Key Laboratory of Pharmacodynamic Constituents of TCM and New Drugs Research, Jinan University, Guangzhou, 510632, China. daixy18@jnu.edu.cn.
  • 6 Institute of Biopharmaceutical and Health Engineering, Tsinghua Shenzhen International Graduate School (SIGS), Tsinghua University, Shenzhen, 518055, China. ma.shaohua@sz.tsinghua.edu.cn.
  • # Contributed equally.
PMID: 41039172 DOI: 10.1038/s44321-025-00308-1
Abstract

Advanced algorithms have significantly improved the efficiency of in vitro screening for protein-interactive compounds. However, target antigen (TAA/TSA)-based drug discovery remains challenging, as predictions of compound-protein interaction (CPI) based solely on molecular structure fail to fully elucidate the underlying mechanisms. In this study, we utilized deep learning, specifically TransformerCPI to screen active molecules from a Chinese herb compound library based on protein sequences. Two natural products, Polyphyllin V and Polyphyllin H, were identified as targeting the pan-cancer marker CD133. Their anti-tumor efficacy and safety were confirmed across validation in Cancer cell lines, tumor patient-derived organoids, and animal models. Despite their analogous structures and binding affinity to CD133, Polyphyllin V suppresses the PI3K-AKT pathway, inducing Pyroptosis and blockage of Mitophagy, whereas Polyphyllin H inhibits the Wnt/β-catenin pathway and triggers Apoptosis. These distinct mechanisms underscore the potential of combining AI-driven screening with biological validation. This AI-to-patient pipeline identifies Polyphyllin V and Polyphyllin H as CD133-targeted drugs for pan-cancer therapy, and reveals the limitations of virtual screening alone and emphasizes the necessity of live model evaluation in AI-based therapeutic discovery.

Keywords

Apoptosis; CD133; Mitophagy; Natural Products; TransformerCPI.

Figures
Products
Inhibitors & Agonists
Other Products