Microsoft Research releases open-source RetroChimera for large-scale small molecule synthesis prediction

Microsoft Research published RetroChimera in Nature, an open-source retrosynthesis model that combines R-SMILES 2 (Transformer) and NeuralLoc (GNN) to automatically propose high-quality synthesis routes for new molecules. This model demonstrates the ability to recall rare reaction types and achieves successful zero-shot transfer and fine-tuning on proprietary datasets, with PhD-level chemists preferring its predictions over preceding models and recorded literature reactions. This addresses the critical, time-consuming, and costly challenge of chemical synthesis in developing new medicines and advanced materials.