RetroBridge: Modeling Retrosynthesis with Markov Bridges

• DOI: 10.48550/arxiv.2308.16212
• Publication Date: 2023-01-01
• Authors: Ilia Igashov; Arne Schneuing; Marwin Segler; Michael M. Bronstein; Bruno E. Correia
• Keywords: Retrosynthetic analysis; Bridge (graph theory)

Retrosynthesis planning is a fundamental challenge in chemistry which aims at designing reaction pathways from commercially available starting materials to target molecule. Each step multi-step retrosynthesis requires accurate prediction of possible precursor molecules given the molecule and confidence estimates guide heuristic search algorithms. We model single-step as distribution learning problem discrete state space. First, we introduce Markov Bridge Model, generative framework aimed approximate dependency between two intractable distributions accessible via finite sample coupled data points. Our based on concept bridge, process pinned its endpoints. Unlike diffusion-based methods, our Model does not need tractable noise sampling proxy directly operates input product samples prior distribution. then address with novel RetroBridge, template-free modeling approach that achieves state-of-the-art results standard evaluation benchmarks.