Researchers from Microsoft Research and collaborators including GSK and Novartis have developed an AI model called RetroChimera to help address the complex task of determining how to make designed molecules in a lab, according to a new study in Nature. The study details how the system performs across public and proprietary chemistry data and how closely its suggestions align with expert chemists’ expectations. Retrosynthesis starts with a desired target molecule and works backward, step by step, to break it down into simpler, commercially available building blocks; its immense combinatorial complexity long led researchers to believe it could not be automated reliably. Paired with a search algorithm, RetroChimera can now propose promising synthesis pathways for target molecules, combining predictions from multiple models with complementary strengths rather than relying on a single AI model. The researchers developed the approach after analyzing common shortcomings in existing retrosynthesis systems, including difficulties incorporating less frequent but strategically important reactions and a tendency to generate inaccurate predictions. The researchers showed that the pre-trained RetroChimera model can be readily adapted to GSK’s internal data and its proprietary chemistry data, suggesting the approach could help apply AI-driven synthesis planning to practical drug discovery challenges rather than only benchmark datasets and academic research tasks, and could make the technology faster and less expensive to customize for other real-world research environments, including small-molecule therapeutics, materials science and fine chemical development.