Human-in-the-Loop Assisted de Novo Molecular Design - 2022
Details
Title : Human-in-the-Loop Assisted de Novo Molecular Design Author(s): Sundin, Iiris and Voronov, Alexey and Xiao, Haoping and Papadopoulos, Kostas and Bjerrum, Esben Jannik and Heinonen, Markus and Patronov, Atanas and Kaski, Samuel and Engkvist, Ola Link(s) : https://doi.org/10.1186/s13321-022-00667-8
Rough Notes
Abstract
This work addresses the bottleneck of how to integrate human feedback to optimize molecules, which involves exploring a large combinatorial chemical space. This is done by introducing a principled probabilistic approach which allows the (chemical) expert to adapt the multi-parameter optimization (MPO) scoring function to better match their goal. This allows for learning the scoring function directly from expert feedback while they explore the output of the molecule generator, rather than the current approach of manually tuning the scoring function with trial and error.
Contributions
- Explicitly model the expert's goal to automatically adapt the scoring function to match their goal.