01 What happened
Microsoft Research and collaborators GSK and Novartis have developed RetroChimera, an AI model that plans retrosynthetic pathways for target molecules. In expert evaluations, it produced fully accepted reaction sequences for nine of ten benchmark molecules, compared with two to five for other models. The approach also adapts to proprietary chemistry data.
02 Key details
- RetroChimera combines multiple AI models with complementary strengths to propose retrosynthetic pathways for target molecules.
- Expert chemists fully accepted its proposed reaction sequences for 9 of 10 benchmark molecules, versus 2 to 5 for prior models.
- Nine Ph.D.-level chemists preferred RetroChimera's top suggestion over documented routes about 64% of the time.
- The pre-trained model adapts to proprietary data, shown with GSK's internal chemistry, enabling faster customization.
- Researchers plan to evaluate RetroChimera in real-world drug discovery settings beyond benchmarks.
03 Why it matters
Retrosynthesis planning is a major bottleneck in drug discovery and materials science. RetroChimera could speed up the process by suggesting feasible synthesis routes that align with expert judgment and work with internal company data, reducing time and cost.
04 Who it matters to
Drug development researchers.
Original source