Prescription drug side effects are one of the most unpredictable aspects of the medical field. There are 5,000 prescription drugs on the market with 1,000 known side effects. Because of this, there are 125 billion possible side effects that could happen when these drugs are combined and doctors often don’t know what the potential drug side effects are when combining prescription drugs. According to a CDC study, in the last month, 23 percent of Americans took two or more prescription drugs and 39 percent of Americans 65 years or older took five or more prescriptions.
Marinka Zitnik and colleagues designed a system to predict billions of potential drug combination side effects. (Source: L.A. Cicero)
The current method for finding out if there are side effects from a combination of prescriptions is to just give them to a patient and wait to see if any side effects occur. This is a dangerous and risky method, but it is the only one available.
"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 Marinka Zitnik, a postdoctoral fellow in computer science. With some new drug combinations, she said, "truly we don't know what will happen."
Researchers have developed a new artificial intelligence (AI) system that hopes to solve this problem. The system is named Decagon. The system can predict the side effects of many drug combinations, help doctors learn more about drug prescriptions and find the best combination for their patients. Decagon is based on a large deep learning network that describes how drugs affect more than 19,000 proteins in the human body.
There are already more than 4 million connections known between drugs and their side effects. Based on this knowledge the team created a method that identified patterns to figure out what side effects may pop up when a new combination of drugs is administered.
In order to test the validity of Decagon’s predictions, the team searched through medical literature for evidence that any of the 10 side effects Decagon predicted actually happened. The team found that five of the 10 side effects were confirmed, proving Decagon to be an accurate prediction system.
Decagon can currently only predict the side effects of a drug pairing. The team wants to further develop the system so it can predict the side effects of more than just two-drug combinations. They also want to develop Decagon to be more user-friendly so it can be used by doctors in clinical settings.
"Today, drug side effects are discovered essentially by accident," Jure Leskovec, an associate professor of computer science, said, "and our approach has the potential to lead to more effective and safer healthcare."
The paper on Decagon was published in Bioinformatics and presented at the International Society for Computational Biology in Chicago.
