TriGlue Generates Molecular Glues for Targeted Protein Degradation

This paper introduces TriGlue, a novel biology-inspired generative model designed to facilitate the discovery of molecular glues. Molecular glues are crucial for targeted protein degradation, a therapeutic strategy that involves forming a ternary complex between an E3 ubiquitin ligase and a target protein. Unlike traditional drug design, molecular glue design is complex, requiring simultaneous modeling of ligand generation, protein-protein docking, and ternary complex assembly. TriGlue tackles this by breaking down the problem into two interconnected stages: first, an SE(3)-equivariant module estimates the geometrically constrained protein-protein interface from unbound monomer structures. Second, an interface-conditioned ternary flow matching network then jointly generates the molecular glue and predicts the necessary rigid-body transformations for assembling the ternary complex. Extensive experiments demonstrate TriGlue's ability to produce chemically valid molecules and plausible ternary complexes, underscoring the potential of this generative modeling approach in accelerating drug discovery.