AI Just made it Possible to Predict the Side Effects of Millions of Drug Combinations

Millions of people around the world are taking multiple drugs on a daily basis and doctors are often ill-informed as to the potential side effects due to the difficulty of predicting the relevant interactions.

Given the plethora of medications on today’s market, “it’s practically impossible to test a new drug in combination with all other drugs, because just for one drug that would be five thousand new experiments,” said Marina Zitnik, a postdoctoral fellow in computer science.

Luckily for patients, the situation is about to change – during the 2018 meeting of the International Society of Computational Biology in Chicago, Zitnik and colleagues introduced a new artificial intelligence system, called Decagon, capable of not only tracking, but also predicting the complex interactions between different pharmaceuticals.

In order to get around the near-insuperable task of analysing the almost 125 billion possible side effects between all possible drug pairs, the research team decided to focus on how individual drugs affect the underlying cellular machinery in the body.

To that end, Zitnik and collaborators drew up a massive network describing how proteins native to our bodies (of which there are close to 19,000) interact with each other and with different pharmaceuticals.

AI predicts side effects of multiple drugs taken simultaneously. Image credit:, CC0 Public Domain.

Using more than 4 million known associations between drugs and side effects, Zitnik’s group deployed a deep learning algorithm to single out specific patterns of interaction based on the manner in which the drugs target different bodily proteins, and predict as yet unseen consequences of taking two individual drugs at the same time.

To verify whether the patterns identified by Decagon actually correlate with real-world results, the researchers cross-referenced a number of side effects predicted by the AI (which did not exist in the original data) with medical literature and found that many of them have been recently confirmed.

In the future, the researchers hope to develop a souped-up, yet user-friendly, version of the system, featuring assessments of more complex regimens, to assist both doctors and medical researchers.

“Today, drug side effects are discovered essentially by accident,” said Jure Leskovec, co-author of the paper out on 27 June in the journal Bioinformatics, “and our approach has the potential to lead to more effective and safer health care.”



Comment this news or article