Abstract

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

Bios:

Arne Schneuing is a PhD student in Bruno Correia’s group at École Polytechnique Fédérale de Lausanne (EPFL) and co-advised by Michael Bronstein. Previously, he obtained a Master’s degree in electrical engineering and robotics from RWTH Aachen and KTH Stockholm. Arne works on geometric deep learning and generative models for the design of molecular interactions between proteins and other biomolecules.

Ilia Igashov is a PhD student at Laboratory of Protein Design and Immunoengineering (EPFL), advised by Bruno Correia and Michael Bronstein, and is a fellow of EPF LglobaLeaders program. His research interests lie in geometric deep learning for biology and chemistry, and especially in applications to protein-protein interactions and drug discovery. Prior to PhD, Ilia did research internships at Inria and Université Grenoble Alpes where he worked on graph neural networks for protein model quality assessment.

Join us on Zoom