Machine learning in targeted protein degradation drug design: a technical review of PROTACs and molecular glues

Abstract

Targeted protein degradation (TPD) allows catalytic removal of disease-associated proteins by exploiting the ubiquitin-proteasome system (UPS). Proteolysis-targeting chimeras (PROTACs) and molecular glues represent two complementary TPD modalities, yet their rational design remains hindered by challenges in ternary complex formation, ligand discovery, and pharmacokinetic optimization. Recent machine learning (ML) advances address these barriers through predictive modeling, virtual screening, and generative design of degrader candidates. In this review, we summarize how ML is integrated across PROTAC and molecular glue development, including ternary complex prediction, linker and fragment design, degradation efficiency modeling, and absorption, distribution, metabolism, excretion, and toxicity (ADMET) optimization. We also highlight emerging artificial intelligence (AI)-driven strategies for de novo glue discovery. Together, these innovations demonstrate how ML is accelerating degrader design and expanding the landscape of druggable targets.

Keywords: PROTAC; deep generative design; machine learning; molecular glue; targeted protein degradation.