A novel deep learning platform in design molecular glues that stabilize protein-protein interactions

Molecular glues can stabilize protein-protein interactions (PPIs) and have emerged as a promising class of therapeutics for many “undruggable” proteins. Deep-learning approaches show enormous potential for revolutionizing structure-based drug design. However, all deep learning methods require a large training dataset containing information on the binding pocket. Such information is often unavailable for molecular glues.

This research project will create a deep-learning platform, Chemical Feature Transformer (CFT), that learns from patterns in chemical environments (using known molecular glue-PPI structures) instead of complementary geometric shapes (like the Key-and-Lock model). To experimentally validate the predictions of the CFT, researchers will apply CFT to identify molecular glues for the degradation of cancer-associated proteins. It is expected that the deep-learning platform can greatly help other UW researchers identify novel molecular glues to tackle their disease-relevant targets. This project will provide the foundation for potentially establishing  a center of excellence in AI-driven molecular design.

PRINCIPAL INVESTIGATOR

Xuhui Huang, professor of chemistry

CO-PRINCIPAL INVESTIGATORS

Sharon Li, assistant professor of computer science

Weiping Tang, professor of pharmacy