Evaluating Retrosynthesis with Syntheseus and Retro-Fallback

Explore a comprehensive lecture on improving synthesis planning evaluation in drug discovery through two innovative projects. Dive into the development of syntheseus, a Python package facilitating easy use and evaluation of established retrosynthesis algorithms. Learn about the "Retro-fallback" approach, which employs a probabilistic formulation of uncertainty and introduces a new search algorithm. Gain insights into the concept of Successful Synthesis Probability and algorithms designed to maximize it. Conclude with a discussion on future directions in the field of synthesis planning, followed by an engaging Q&A session. This 58-minute talk, presented by Austin Tripp from Valence Labs, offers valuable knowledge for professionals and researchers in the AI for drug discovery community.

- Intro + Background
- Syntheseus
- Retro-fallback
- Successful Synthesis Probability
- Algorithms to Maximize SSP
- Conclusions
- Q&A