At a time where very few people live without some or the other pharmaceutical product on a daily basis, the industry spoils us with a choice of generic medicines for every ailment. Ultimately, though, the choice of one over dozens of similar medications falls in the hands of the consumer, and so does its safety.
While every approved drug goes through the arduous process of human trials, it is next to impossible to test them in combination with every other drug that might be consumed alongside it. If done, it would be a wildly impractical experiment using more resources than the average laboratory has at its disposal.
Many of the drug interactions that are known today have been discovered accidentally, and over long periods of trial-and-error. The sole defence that doctors, pharmacists and consumers have used till date against side effects from these interactions has been luck and chance. Enter Artificial Intelligence.
An AI designed by Stanford computer scientists successfully predicted " and not just tracked " side effects from millions of possible drug interactions. The team published >a study with their research and findings from the AI, which they've decided to call 'Decagon'. They got around the complexity of identifying unwanted drug interactions by going back to the fundamentals of how drugs affect a cell's machinery.
Decagon's makers are Dr Marinka Zitnik, a postdoctoral fellow, Monica Agrawal, a master's student, and Dr Jure Leskovec, an associate professor of computer science at Stanford University.
The trio mapped out a comprehensive network of the nearly 20,000 proteins in our body, along with what we know about how they interact with each other, and finally included the nearly 4 million known interactions these proteins have with pharmaceuticals.
This made the process of finding patterns of why and how a certain drug causes a protein to behave differently. This gargantuan task was outsourced to Decagon, which used deep learning" an AI instrument that crunches complex data and what could be eliminated as counter-intuitive patterns by people" to find every possible contraindication it can for a chosen drug cocktail.
The team also explained that Decagon finding these patterns doesn't make it absolutely certain, but most of its predictions that they tested afterward were bang-on. Citing an example from their presentation, the team shared how Decagon predicted muscle inflammation from a combination of atorvastatin (a cholesterol drug) and amlopidine (a blood pressure drug), that AI didn't foresee. The team cross-referenced similar predictions by Decagon with medical literature and found 50 percent of them confirmed by recently published studies.
While the AI currently processes predictions from drug pairs, the team is working towards extending its abilities to complex regimens akin to those in the real world.
Once modified to make it more user-friendly for non-computer scientists, the team sees huge potential for Decagon in helping doctors make better decisions about what drugs to prescribe, and fellow researchers in finding better drugs combinations for complex illnesses.