ETH AI model helps to find new drugs

Scientists from the Chemistry and Artificial Intelligence research departments at ETH Zurich, together with experts from Roche Pharma Research and Early Development, have developed a process that helps to find the optimal manufacturing process for new drugs, according to a press release. They work at the interface between academic AI research and laboratory automation.

Whereas previously possible pathways for laboratory production were derived from known chemical reactions and then individually tested experimentally in a complex trial-and-error process, their new AI model predicts the suitable molecular sites. Borylation was chosen as the chemical activation method for the underlying active substance scaffold.

In a comprehensive worldwide literature search, the team led by PhD student Kennetz Atz and Professor Gisbert Schneider from the Institute of Pharmaceutical Sciences found 38 particularly trustworthy scientific papers on boronylation, which described a total of 1380 boronylation reactions. This training data set was supplemented by 1000 reactions carried out in the automated laboratory of Roche's medicinal chemistry research department.

The AI model generated from this data pool not only predicts the best locations, but also indicates where no activation is possible in the scaffolding and provides the optimum conditions for the activation reactions in each case. In the same way, the team now wants to find effective models for activation reactions other than borylation. In addition, Atz is now involved in medicinal chemistry research at Roche as an AI scientist. "This innovative project is a successful bridge between academia and industry," says Schneider. It demonstrates "the enormous potential of public-private partnerships for Switzerland". ce/mm