Ask Onix
AI transforms the search for new medicines
Artificial intelligence is rapidly reshaping drug development, offering fresh hope against antibiotic-resistant infections, Parkinson's disease and thousands of rare conditions that have long eluded treatment.
The antibiotic crisis
Global health faces a growing threat from drug-resistant bacteria. Over 1.1 million people die annually from infections once easily cured, with projections warning of eight million deaths by 2050 without intervention. Traditional antibiotic development has stalled, with only 12 new drugs approved between 2017 and 2022-most resembling existing treatments that bacteria are already evading.
James Collins, a professor at MIT, explains how AI is changing the game. His team trained generative models to recognize chemical structures of known antibiotics, then screened over 45 million compounds for potential effectiveness against Neisseria gonorrhoeae and MRSA-both notorious for resisting current drugs. The process identified two promising new compounds that attack bacteria differently than existing antibiotics, now undergoing further testing.
"We can analyze massive libraries of chemical compounds in days or hours-something that would take years with conventional methods," Collins said.
Tackling Parkinson's with machine learning
Parkinson's disease affects over 10 million people worldwide, with no treatments to slow its progression. Michele Vendruscolo at the University of Cambridge turned to AI after decades of failed clinical trials left researchers uncertain about the disease's origins. His team used machine learning to identify compounds targeting Lewy bodies-protein clumps linked to neurodegeneration in Parkinson's patients.
The AI screened billions of molecules, narrowing the field to five promising candidates. Unlike traditional methods that might test a million compounds over six months, Vendruscolo's approach achieved similar results in days at a fraction of the cost. The compounds are now in further testing, with hopes they could one day halt the disease before symptoms appear.
Repurposing existing drugs
AI isn't just creating new drugs-it's also finding new uses for old ones. David Fajgenbaum, a University of Pennsylvania professor, used an existing drug to treat his rare Castleman disease after conventional therapies failed. His nonprofit, Every Cure, now employs machine learning to match thousands of approved drugs with potential new applications.
Researchers at Harvard Medical School identified nearly 8,000 drugs that could treat 17,000 diseases, while a McGill University team used AI to repurpose a hypertension medication for idiopathic pulmonary fibrosis (IPF), a rare lung disease. Their model simulates disease progression, allowing researchers to test treatments virtually before moving to clinical trials.
Challenges and limitations
Despite these advances, AI-driven drug discovery faces hurdles. Many critical datasets remain locked within pharmaceutical companies, limiting AI's ability to predict drug properties like toxicity and absorption. While AI excels at early-stage screening, the full drug development process remains lengthy and uncertain.
"AI is revolutionizing drug discovery, but only in very specific ways," Vendruscolo noted.
Companies like Insilico Medicine are pushing boundaries, with their AI-designed drug Rentosertib showing promise in IPF clinical trials. If successful, it could reach patients by the end of the decade. Still, experts caution that widespread AI-driven drug development may take years to materialize.
The future of AI in medicine
As AI tools grow more sophisticated, researchers predict they will play an increasing role in drug discovery. Jun Ding of McGill University believes most new drugs could be AI-guided within the next decade. For now, the technology offers a powerful new weapon in the fight against diseases once considered untreatable.
