Izaguirre, J. A.; Wu, Y.; McDargh, Z.; Palpant, T.; Razavi, A. M.; Trovato, F.; Koh, C.; Xu, H.

GlueMap is a computational toolkit to discover and optimize molecular glues. The tool integrates structural and pharmacodynamic modeling, molecular dynamics (MD), and machine learning to model targeted protein degradation (TPD) mediated by molecular glues (MGs). GlueMap effectively models ternary complexes, predicts their thermodynamic stabilities, and evaluates downstream effects like ubiquitination and degradation efficiency. We validated GlueMap\'s capabilities through two case studies: CRBN-dependent degradation of GSPT1 and DDB1-dependent degradation of CDK12. For CRBN-GSPT1, GlueMap successfully recovered known molecular glue degraders and identified novel candidates through prospective virtual screening using ternary complexes derived from extensive MD simulation. In the DDB1-CDK12 system, we observed strong correlations between ternary complex stability, quantified by binding free energy calculations, and degradation efficiency, though notable exceptions highlighted the importance of considering structural dynamics of the ternary complex in the context of the cullin ring ligase. Through a supervised variational autoencoder (VAE) model combined with attention-based regression, GlueMap demonstrates high accuracy in predicting degradation potency from structural features. By integrating structural, thermodynamic, and ubiquitination metrics, GlueMap provides a multi-criteria toolkit for rational MG design, distinguishing it from traditional screening methods that rely solely on docking scores. These results illustrate that GlueMap may provide mechanistic insights and accelerate molecular glue discovery and optimization across diverse protein targets.