AI predicts drug interactions

Stanford (pte002 / 11.07.2018 / 06: 05) –

Researchers at Stanford University http://stanford.edu have developed "Decagon", an AI system to predict side effects when taking two drugs. Because for most combinations, potentially dangerous interactions are so far unknown. The system could help make the treatment of serious diseases safer.

Risky Cocktails

More and more people are taking more and more medicine. Older patients often receive a badtail of whole pills daily. However, with thousands of approved drugs, it is virtually impossible to test the interactions of all possible combinations. Especially new combinations are risky. "We really do not know what's going to happen," says Marinka Zitnik, a postdoctoral fellow at Stanford. That's what her colleagues and she want to do with Decagon

The team has modeled how more than 19,000 proteins interact in the body and how drugs affect them. By using four million known drug combinations and side effects, they have come up with a method for detecting side-effect profiles of the effects of drugs on proteins. For this, the team is put to the deep learning. The system is designed to predict first the side effects for two drug combinations of concomitant use.

Predicting the Unexpected

The team has a number of side effects that Decagon predicted but did not appear in the original data set. Check if they are now included in the medical literature. The researchers found that the AI ​​system warned of dangerous and unexpected muscle inflammation by taking the antihypertensive drug amlodipine and the Atorvastatin cholesterol drug. In addition, five of Decagon's other ten predictions have shown that side effects have now been proven in practice.

This suggests that the AI ​​approach is indeed likely to predict potential drug interactions in a relatively reliable manner. The researchers hope to extend the system to combinations of three or more drugs. In addition, they want to develop a user-friendly tool that doctors can use. The accidental detection of potentially serious side effects on patients could then come to an end. "Our approach has the potential to provide more effective and safer health care," says Jure Leskovec, a professor of computer science at Stanford.

(end)