Advertisement

With 125 billion possible side effects between all possible pairs of drugs, accurately predicting how a patient may react to a new drug can be a dangerous guessing game.

However, a team from Stanford University is using a new artificial intelligence (AI) system to better predict potential side effects from drug combinations.

The new system, dubbed Decagon, could aid doctors when prescribing drugs to patients already on a laundry list of medications, while also assisting researchers in finding better combinations of drugs to treat complex diseases.

There are 1,000 different known side effects and 5,000 drugs on the market—the majority of which have never been prescribed together. This makes determining the impact of different drug combinations difficult.  

“It's practically impossible to test a new drug in combination with all other drugs, because just for one drug that would be 5,000 new experiments,” Marinka Zitnik, a postdoctoral fellow in computer science, said in a statement.

The researchers began studying how drugs affect the underlying cellular machinery in the body. They composed a substantial network that describes how the more than 19,000 proteins in the body interact with one another and how different drugs affects those proteins.

They used more than four million of the 125 billion known associations between drugs and side effects to design a deep learning system that identifies patterns in how side effects arise based on how drugs target different proteins.

The deep learning system infers patterns about drug interaction side effects and then predicts what the consequences from taking multiple drugs together would be.

The researchers then set out to confirm whether the predictions were accurate.

For example, despite no previous indication that the cholesterol drug atorvastatin combined with the blood pressure drug amlipidine would lead to muscle inflammation, the AI system predicted it would. The researchers later confirmed after finding a 2017 case report that suggested that the drug combination led to a dangerous kind of muscle inflammation.

They also found at least 10 side effects predicted by Decagon that were not in their original data, half of which having been recently confirmed.

“It was surprising that protein interaction networks reveal so much about drug side effects,” Jure Leskovec, an associate professor of computer science, said in a statement.

The researchers next plan to build on their work and include more complex regimens, while also creating a more user-friendly tool for doctors.      

According to the Centers for Disease Control and Prevention, 23 percent of Americans took at least two prescription drugs last month and 39 percent of people over 65 in the U.S, took five or more.

 

Advertisement
Advertisement