Microsoft Developed AI for Chemical Synthesis
Researchers launched Retrochimera in 2025 to improve retrosynthetic analysis through a modular AI system.
Updated on Oct. 13, 2026 in Chemistry
Live Poll
Do you trust AI-powered tools to improve accuracy in technical and scientific research?
Released in 2025, Microsoft's new Retrochimera tool performs retrosynthetic analysis by cross-checking two distinct neural networks. The system aims to simplify complex chemical planning by reducing errors common in individual AI models.
Why it matters
Combining different AI models reduces the failures and hallucinations often seen in previous retrosynthesis tools. This modular approach allows for better accuracy in identifying chemical reaction steps.
The tool successfully reduced AI hallucinations by more than half on industrial reactions during testing. It was validated using 10,444 reactions from Novartis and 650,000 reactions provided by GSK.
The players
Microsoft
Microsoft is a global technology corporation that maintains a dedicated research division focused on AI and scientific computing.
Novartis
Novartis is a Swiss multinational pharmaceutical company that provided data for the testing of the new AI system.
GSK
GSK is a British pharmaceutical and biotechnology company that contributed hundreds of thousands of reactions to support model training.
The details
Retrochimera utilizes a transformer-based submodel called R-Smiles 2 and a graph neural network called Neuralloc to analyze input molecules. By outputting a ranked list of reactions, the tool consistently outperformed its individual components, particularly regarding rare chemical transformations.
Timeline
Retrochimera was released in 2025.
This report was published on October 13, 2026.
The Big Picture
Retrochimera represents a significant shift toward modular AI design in computational chemistry, directly addressing the reliability issues of earlier retrosynthetic software. This development marks a departure from monolithic models, potentially setting a new standard for how AI architecture is used in chemical synthesis research.
The ability of this tool to reduce AI hallucinations could lead to more efficient and reliable methods for developing new medications. By streamlining the planning of complex chemical reactions, researchers may see faster timelines for bringing new chemical compounds from the laboratory to industrial application.
The takeaway
Reliability in AI-driven chemistry is best achieved by cross-referencing multiple specialized models to catch errors early. Chemists looking to adopt new tools should prioritize systems that allow for fine-tuning against proprietary reaction databases.
Further reading
For more on the latest computational advancements in the field, visit Chemistry.
Source note: This article includes information reported by Chemistry World.
Live Poll
Do you trust AI-powered tools to improve accuracy in technical and scientific research?
