Palo Alto, CA – Complications from drug interactions can present patients with whole new afflictions. The concern is widespread, with 39% of America’s senior citizens taking five or more prescription drugs simultaneously. It’s surprising, therefore, how little is known about potential interactions between many drugs, even some commonly prescribed ones.
The main reason so much here remains unknown is that, with the number of prescription drugs currently available, polypharmacy (the practice of combining drugs) may have the potential for more than 125 billion complications. Testing for all of them would be nearly impossible; most interaction side effects are discovered by accident. Accordingly, in many cases, when a doctor adds another pill to a patient’s regimen, they have no clue what complications might occur.
But a brand-new AI system will not only equip healthcare providers with whole new bodies of knowledge about drug interactions—most notably, it will arrive at that knowledge on its own. Developed at Stanford University, “Decagon” models the human body’s protein networks: the thousands of proteins that interact with drugs and with one another. Once Decagon was fed the knowledge we already have about pharmaceutical interactions, it was able to identify patterns and use them to make predictions about other drug combinations.
Protein networks aren’t an obvious window onto drug interactions, but Decagon has used them to promising effect: the system beat baseline conjectures by almost 70%, accurately predicting complications from many novel combinations of drugs. For example, it told researchers that taking the blood-pressure medication amlodipine alongside cholesterol-reducing atorvastatin could cause muscle inflammation. It turned out to be right.
As of now, Decagon is capable of predicting interactions only within two-drug combinations. Its creators hope that, in the future, it will be able to tackle more complicated regimens.
Why This Matters
While other AI systems are already revolutionizing the realm of patient medication—by dispensing established guidance when appropriate—Decagon, with its independent predictions, equips the healthcare community at large with guidance entirely new. It can arm HCPs with novel awareness and help ensure safe courses of treatment for patients. As Dr. Jure Leskovec, one of Decagon’s creators, says, “Our approach has the potential to lead to more effective and safer healthcare.”
Also noteworthy is the amount of money that might be saved on drug research and development. According to a McKinsey estimate, the big data offered by AI systems could save pharmaceutical companies up to $100 billion a year.
