Microsoft’s split-personality retrosynthetic AI is a fast learner

By Andy Extance2026-10-13T13:30:00

Example of a retrosynthesis tree. For a single target molecule, many disconnections are possible, which introduces a high branching factor. The figure shows many incomplete routes in pale colors that contrast a completed route, which connects all the way

Source: © Microsoft

Retrochimera combines two deep learning algorithms from different classes, outperforming their individual success rates on datasets from Novartis and GSK

Bringing together two different AI models built to ‘think’ in entirely different ways reduces failures and hallucinations that can arise when they perform an important chemical task – retrosynthesis.

‘We find that different models have different strengths and that combining them can yield considerable performance improvements over any single model alone,’ says Marwin Segler, senior director at Microsoft Research AI for Science in Cambridge, UK. The combined tool, Retrochimera, can also be fine-tuned cheaply and quickly to work with new databases of reactions.