EPFL Researchers Launch Synthegy AI Framework to Revolutionize Molecular Design Through Natural Language Reasoning

EPFL researchers unveil Synthegy, an AI framework that uses natural language to help chemists plan molecular synthesis and reaction mechanisms more efficiently.

By: AXL Media

Published: Apr 25, 2026, 9:33 AM EDT

Source: Information for this report was sourced from EurekAlert!

Bridging the Gap Between AI and Human Chemical Intuition

The traditional process of molecular design has long been hampered by a disconnect between vast computational search capabilities and the nuanced strategic reasoning of expert chemists. While computers can efficiently navigate massive chemical spaces, they frequently lack the high-level foresight required for complex retrosynthesis. A team at EPFL has addressed this limitation by developing Synthegy, a new framework that positions large language models as reasoning engines rather than simple generators. This shift allows the AI to evaluate and steer traditional algorithms using the same logic and specialized vocabulary employed by human researchers.

Streamlining Retrosynthesis with Natural Language Interfaces

One of the primary hurdles in modern chemistry is retrosynthesis, the process of working backward from a target molecule to identify simpler building blocks. Synthegy simplifies this task by enabling chemists to provide instructions in plain language, such as requesting the early formation of a specific ring or the avoidance of unnecessary protecting groups. The system then translates potential synthetic routes into text, allowing the language model to score and explain the logic behind each pathway. According to first author Andres M Bran, this interface removes the need for cumbersome manual filters, allowing scientists to iterate much faster.

Unraveling Complex Reaction Mechanisms Through Electron Movement

Beyond planning synthesis, the Synthegy framework provides critical insights into reaction mechanisms by breaking them down into elementary electron movements. The system explores multiple potential pathways and utilizes AI to identify the most chemically plausible options based on specific reaction conditions or expert hypotheses. This mechanistic clarity helps researchers predict new reactions and improve overall efficiency, reducing the reliance on costly and time-consuming trial and error. By integrating these two distinct problems into a single interface, the technology offers a unified approach to molecular discovery.