Microsoft researchers released RetroChimera to simplify drug synthesis planning
The open-source AI model combines two different approaches to accurately predict how to create complex molecules in a lab.
Microsoft researchers published a study in the journal Nature introducing RetroChimera, an AI framework designed to plan the synthesis of small molecules. This tool helps chemists determine the step-by-step process required to build new medicines and materials, which is typically a manual and expensive task.
The system functions by evaluating two distinct models. R-SMILES 2 acts as a transformer-based generator that proposes precursors, while NeuralLoc uses graph neural networks to match chemical templates. By ranking the outputs of both, the software identifies reliable reaction routes that experts in blind tests preferred over existing methods.
Researchers made the code and model weights available on GitHub and through Microsoft Foundry. The team hopes these resources will assist laboratories in automating the design-make-test cycle for chemical research.
RetroChimera relies on patterns learned from previous data. Its effectiveness is limited by the diversity of its training set, as it remains constrained when attempting to predict reactions that fall outside of its established patterns.
Source: Microsoft Research Blog