Generative AI reshapes the PROTAC discovery paradigm: From empirical screening to rational design
TL;DR
This perspective explores how multimodal deep learning enhances ternary complex prediction accuracy and boosts molecular design hit rates from 5-8% to over 35%.
Abstract
PROTACs are revolutionary therapeutics that eliminate pathogenic proteins via catalytic event-driven degradation. Conventional development suffers from unpredictable ternary complexes, empirical linker modification and blind ligand screening, yielding clinical success below 10% and requiring 3-4 years to obtain preclinical candidates. Generative artificial intelligence (AI) offers a promising avenue for data-driven rational design. This perspective explores how multimodal deep learning enhances ternary complex prediction accuracy and boosts molecular design hit rates from 5-8% to over 35%. Closed-loop research and development (R&D) ecosystems supported by databases accelerate PROTAC development. The landmark case of Insilico Medicine’s PKMYT1-PROTAC, developed via its AI platform Chemistry 42, achieves a DC50 of 0.5 nM and exhibits a dual degradation-inhibition mechanism, and has advanced to the preclinical candidate (PCC) validation stage within approximately 12 months. We further discuss how to integrate AI design with clinical translation. Future efforts focus on advanced computational strategies, establishing standardized evaluation metrics and expanding the E3 ligase toolbox. This AI-powered strategy will expedite PROTAC research and expand the application of targeted protein degradation therapy.