Small molecules

We work on accelerating the discovery of small molecules through AI at all steps of the Design-Make-Test cycle. This includes models to predict molecular properties, generative models to design new molecules with the required properties, and models to predict how molecules can be synthesized.

RetroChimera

RetroChimera is a predictive model designed to assist in chemical synthesis. Given a target product molecule, either represented as molecular graphs or as a sequence of characters (SMILES), it generates multiple plausible chemical reactions that could produce the desired compound. Each reaction consists of a set of reactant molecules, represented either as molecular edits or as character strings, generated de novo.
The model is intended to support the synthesis of drug-like small molecules and is being shared with the research community to encourage reproducibility and stimulate further exploration in this domain.
RetroChimera is aimed at domain experts who possess the expertise to critically assess the quality of its outputs before applying them in practice.

Syntheseus

Syntheseus is a modular Python library for retrosynthetic planning that constructs synthesis routes for target molecules by repeatedly applying reaction prediction models and assembling the results into reaction trees. It supports integration with various reaction models and search algorithms, making it a flexible tool for automated chemical synthesis planning.

MoLeR

MoLeR is a generative model for molecular graphs based on a variational autoencoder, enabling the creation of novel molecules with specified substructures or scaffolds. It is designed to support scaffold-constrained generation and exploration of chemical diversity for applications like drug discovery.