Buffalo, New York: Treating cancer is rarely as simple as prescribing a single medicine. But finding the right mix of drugs and treatment is a slow, expensive and uncertain process, involving years of laboratory work and trial-and-error testing. So, researchers are now turning to Artificial Intelligence to speed that up.
A study published in Nature Communications in April explored whether machine learning models could predict the most effective drug combinations for pancreatic cancer — one of the deadliest and hardest-to-treat forms of the disease.
Researchers first screened thousands of compounds in the laboratory and
shortlisted the most active ones. They then asked three independent teams from different institutes to use separate machine-learning models to predict the best drug combinations.
When the top-ranked combinations were tested, nearly 60 per cent showed improved anticancer activity. The findings suggest that AI can sharply reduce the amount of preclinical testing required, replacing much of the costly trial-and-error approach traditionally used in drug discovery.
The study is significant because developing even a single new drug is enormously expensive. A 2024 report in JAMA Network estimated the median cost of bringing one drug to market is roughly $879.3 million. Combination therapies can cost even more, since each pairing requires additional testing, safety studies and clinical trials, a 2014 BCG report on the pricing of combination therapies found.
Also Read: Cancer therapy may soon skip the lab step — and cut costs
Tech still evolving
AI researchers say machine learning could help narrow down the most promising candidates far earlier in the process, potentially saving years of work and millions of dollars. But scientists caution that the technology is still evolving.
Different AI models often produce partially conflicting rankings for the same drug combinations. Those differences can stem from how the systems are trained, the type of biological data they use, and even how they define “synergy” — the degree to which two drugs work better together than alone.
That means a model trained for one cancer may not necessarily work well for another. Researchers say AI tools will need to be carefully tailored and validated for specific diseases before they can be reliably used in clinical oncology.
Similar attempts have shown encouraging results in the past. A 2019 study published in Nature Machine Intelligence described a machine-learning model DECREASE, designed to reduce the experimental burden of screening drug combinations.
Researchers found that the system successfully identified several highly synergistic drug pairs, four of which later advanced to clinical trials. Two have completed Phase 1 testing, while two others are currently in Phase 2 trials.
But scientists argue the long-term payoff could extend beyond cost savings. If refined further, AI-driven systems may eventually help doctors tailor drug combinations to the exact mutation profile of an individual patient’s cancer, bringing oncology closer to truly personalised treatment.
Arun Singh is an intern at ThePrint. He is an alumnus of ThePrint School of Journalism.
(Edited by Stela Dey)