Credit: Michael S. Helfenbein with AI-generated images

Speeding up drug discovery in the age of AI may come down to a concept that’s comfortingly old-fashioned: Consulting a chemistry recipe book.

It makes perfect sense. Designing a new synthetic molecule has always been a daunting, time-consuming endeavor—made even more challenging by the sheer amount of new research by scientists worldwide. Nearly every week there are innovative new discoveries, protocols, best practices, and shortcuts that can be brought to bear on any given step in a new chemical synthesis, assuming a researcher is aware of them.

How MOSAIC leverages AI expertise

Chemists at Yale, working with researchers from the US unit of Boehringer Ingelheim Pharmaceuticals in Connecticut, have developed an AI-powered platform of digital "experts" to provide just such a guide. Called MOSAIC, it is an AI framework that generates experimental procedures for chemical synthesis, including for compounds that don’t currently exist.

"Chemistry has accumulated millions of reaction protocols, but making practical use of that knowledge remains a bottleneck," said Yale’s Victor Batista, who led the research and is co-corresponding author of a new study in the journal Nature. "MOSAIC is designed to transform that information overload into actionable laboratory procedures."

Batista is the John Gamble Kirkwood Professor of Chemistry in Yale’s Faculty of Arts and Sciences, a member of the Energy Sciences Institute and the Yale Quantum Institute, and director of the Center for Quantum Dynamics on Modular Quantum Devices.

Advantages over traditional AI models

Batista and his colleagues say MOSAIC outperforms other AI-aided resources for chemistry because it is powered by 2,498 individual AI "experts"—each one representing the knowledge of a leading practitioner in a particular chemistry-related topic.

It is akin to cooking a meal by simultaneously consulting with the world’s best chefs for making a roux, choosing spices, and knowing what precise temperature to use.

"Chemists follow recipes to synthesize molecules, just like chefs follow recipes from a cookbook," said Timothy Newhouse, a professor of chemistry at Yale and co-corresponding author of the study. "Being able to quickly look up protocols to make molecules with MOSAIC makes synthetic chemistry easier, just like ChatGPT has made finding a fun new recipe easier."

The first authors of the study are Haote Li, a member of Batista’s lab who earned his Ph.D. in chemistry at Yale in 2025, and Sumon Sarkar, a postdoctoral fellow in Newhouse’s lab.

Real-world impact and future potential

The researchers noted that existing AI chemistry systems rely on a single, large model to assist users. The MOSAIC framework, they said, allows users to cull expertise from thousands of distinct niches of chemical reactions.

"We demonstrated in this work that such an approach outperforms commercial large language models on similar tasks while realizing vast compounds across truly diverse chemical spaces, including pharmaceuticals, catalysts, advanced materials, agrochemicals, and even cosmetics products," Li said.

Indeed, the Yale team used MOSAIC to successfully synthesize more than 35 previously unreported compounds.

The MOSAIC framework is also designed to provide users with measurable uncertainty estimates—reflecting how closely a request fits into a MOSAIC "expert’s" domain of experience—which will enable users to prioritize their experiments.

The new system is fully open-source and compatible with future models likely to emerge, the researchers said. They added that MOSAIC is intended to help move AI beyond prediction and more directly into supporting real-world experimentation.

"Chemistry has evolved from books to databases, and now to AI-guided navigation," said Sarkar. "At a high level, MOSAIC functions like a smart cookbook for new recipes and Google Maps for navigating chemical synthesis. It helps chemists turn vast knowledge into detailed, reproducible procedures for synthesis with an indication of how likely they are to work."

Publication details

Li, H et al, Collective intelligence for AI-assisted chemical synthesis. Nature (2026). DOI: 10.1038/s41586-026-10131-4. www.nature.com/articles/s41586-026-10131-4

Journal information: Nature

Citation: MOSAIC platform compiles chemistry protocols for faster drug design (2026, January 19) retrieved 19 January 2026 from https://phys.org/news/2026-01-mosaic-platform-chemistry-protocols-faster.html

This document is subject to copyright. Apart from any fair dealing for the purpose of private study or research, no part may be reproduced without the written permission. The content is provided for information purposes only.