MIT researchers introduced BoltzGen, an open-source generative AI system that designs protein binders for previously “undruggable” disease targets, potentially accelerating drug discovery. The model unifies protein design and structure prediction, bakes in chemistry- and physics-aware constraints, and was validated across 26 targets in eight academic and industry wet labs. Its release follows earlier Boltz models and underscores a shift from understanding biology to engineering it, the team said. Industry observers note the open-source pace could pressure commercial “binder-as-a-service” startups as performance gaps narrow. The work is led by PhD student Hannes Stärk, with senior authors Regina Barzilay and Tommi Jaakkola, and arrives amid broader efforts to apply AI to biomedical R&D.
Related article:
