RetroChimera AI improves chemical synthesis planning

Microsoft Research has developed RetroChimera, an artificial intelligence system designed to improve chemical retrosynthesis planning. The tool aims to solve the bottleneck in manufacturing complex molecules by suggesting plausible sequences of chemical transformations, from a target molecule back to available precursors.

RetroChimera utilizes a dual-model architecture to identify patterns in data more effectively than single-model systems. In rigorous testing, the AI successfully resolved nine out of ten complex synthesis objectives presented to expert chemists. This performance significantly outperformed existing benchmarks, such as NeuralSym, which resolved only two out of ten.

In blind tests, PhD-level chemists preferred the disconnections suggested by RetroChimera over those from previous systems and, in some instances, over reactions documented in scientific literature. While the tool is not intended to replace laboratory work, it serves as a decision-support system to help specialists evaluate accessible reagents and reasonable reaction conditions.