End-to-end sequence-structure-function meta-learning predicts genome-wide chemical-protein interactions for dark proteins

Tian Cai, Li Xie, Shuo Zhang, Muge Chen, Di He, Amitesh Badkul, Yang Liu, Hari Krishna Namballa, Michael Dorogan, Wayne W. Harding, Cameron Mura, Philip E. Bourne, Lei Xie

Abstract: Author summary Many complex diseases, such as Alzheimer’s disease, mental disorders, and substance use disorders, do not have safe and effective therapeutics because of the polygenic nature of the diseases and a lack of thoroughly validated drug targets (and their corresponding ligands). Identifying small-molecule ligands for all proteins encoded in the human genome would provide powerful new opportunities for drug discovery of currently untreatable diseases. However, the small-molecule ligand of more than 90% of gene families is completely unknown. Existing protein-ligand docking and machine learning methods often fail when the protein of interest is dissimilar to those with known functions or structures. We have developed a new deep learning framework, PortalCG, for efficiently and accurately predicting ligands of understudied proteins which are out of reach of existing methods. Our method achieves unprecedented accuracy versus state-of-the-art approaches, and it achieves this by incorporating ligand binding site information and the sequence-to-structure-to-function paradigm into a novel deep meta-learning algorithm. In a case study, the performance of PortalCG surpassed the rational design from medicinal chemists. The proposed computational framework can shed new light on how chemicals modulate biological systems, which is indispensable in drug repurposing and rational design of polypharmacology. This approach could offer a new way to develop safe and effective therapeutics for currently incurable diseases. PortalCG can be extended to other types of tasks, such as predicting protein-protein interactions and protein-nucleic acid recognition.