People take a lot of drugs: According to the CDC, about a quarter of us took two prescription drugs in the last month, and people over 65 routinely often take five or more — some as many as 20. And the thing is, doctors don't always know what side effects some combinations might cause — there are just too many combinations to test. What we know about those side effects is largely discovered by accident and recorded in adverse-event reporting systems.
Fortunately, computer scientists may be able to help. As I explain in a Stanford News story, researchers Marinka Zitnik, PhD; Monica Agrawal; and Jure Leskovec, PhD, have designed a new system to deal with the literally billions of possibilities when considering any two drugs — out of about 5,000 on the U.S. market — and one of around a thousand different side effects. Their system builds a partial picture of those possibilities, starting with what data is already available about how proteins in our bodies interact with drugs and what adverse side effects have already been reported.
Then, the system known as Decagon fills in the rest using artificial intelligence — and it works. In one case, it predicted a side effect of taking a cholesterol drug with a blood pressure medication that has only recently been reported in the medical literature. Although there are some limitations, the researchers said their idea could make a big difference in the clinic:
'Today, drug side effects are discovered essentially by accident,' Leskovec said, 'and our approach has the potential to lead to more effective and safer health care.'
Zitnik and Leskovec presented their research this week at the 2018 meeting of the International Society for Computational Biology, and a version of the paper appeared in Bioinformatics late last month.
Photo of Marinka Zitnik by L.A. Cicero