Diffusion Models for Sampling Rare Events in Chemistry

Explore the cutting-edge application of diffusion models in sampling rare events, specifically focusing on transition state search in chemistry. Learn how an object-aware SE(3) equivariant diffusion model can revolutionize the process of generating accurate 3D transition state structures for elementary reactions. Discover how this innovative approach significantly reduces computational time from hours to seconds, while maintaining high accuracy. Delve into the model's ability to satisfy physical symmetries and constraints, its performance in generating transition state structures, and its potential for constructing large reaction networks with unknown mechanisms. The talk covers key topics including background information, DDPMs for molecule generation, object-wise SE(3) symmetry, transport, and concludes with a summary and Q&A session. Gain valuable insights from speaker Chenru Duan in this 48-minute presentation from Valence Labs, a leader in AI for drug discovery.