AI CHEMISTRY
Turbo-boosting chemistry with AI
By joining forces, EPFL and Roche have applied artificial intelligence to optimising chemical synthesis in a way that is much more effective than traditional experimental methods.
This laboratory robot can carry out 96 experiments in parallel. | Photo: F. Hoffmann-La Roche
Serendipity’s days may be numbered. Until recently, optimising chemical reactions depended heavily on scientists’ instincts. Today, artificial intelligence is making its way into laboratories. The Minerva system – developed by an international scientific team affiliated with EPFL and the company F. Hoffmann-La Roche – has enabled the guidance of laboratory robots and the performance of 96 parallel reactions, without leaving anything to chance.
Managing so many automatic platforms was the first success. “It showed the potential of machine learning in a high-speed experiment laboratory”, says Philippe Schwaller, a professor at the Laboratory of Artificial Chemical Intelligence (LIAC) at EPFL. Thanks to Minerva, reactions have been optimised where chemists following their intuition were stumped using the traditional experimental method. In another case, an optimisation took four weeks instead of six months.
“This way of working with Roche is really exemplary, says Schwaller. The code, the reactions, the iterations and the measurements are all freely accessible. The data are also useful for retraining new models and validating other techniques”.
Yield and selectivity were prioritised in the first tests. But Minerva can optimise many other factors, including sustainability and the environmental impact of processes. “There’s huge potential for pharmaceuticals, and also for many other industries!”.
