Artificial intelligence is accelerating drug discovery for diseases that have resisted treatment for centuries, from drug-resistant bacteria to Parkinson's disease, by analyzing millions of chemical compounds in hours instead of years. Researchers are using machine learning to identify entirely new compounds that could overcome bacterial resistance and target the root causes of neurodegenerative diseases, marking a fundamental shift in how we approach some of medicine's most stubborn problems. Why Are Drug-Resistant Infections Becoming a Crisis? For roughly 50 years, humanity has been losing ground against bacteria. Antibiotics, once our most powerful weapons against infection, are becoming increasingly ineffective as drug resistance spreads globally. Around 1.1 million people die every year from infections that were easily treatable just decades ago. Without urgent intervention, that death toll is expected to climb to more than 8 million annually by 2050. The problem is compounded by a broken drug development pipeline. Between 2017 and 2022, only 12 new antibiotics were approved for use, and most of them were similar to existing drugs that bacteria are already learning to resist. Pharmaceutical companies have largely abandoned antibiotic research due to low profit margins, leaving the field chronically underfunded and understaffed. How Is AI Changing Antibiotic Discovery? Enter artificial intelligence. Researchers at the Massachusetts Institute of Technology, led by Professor James Collins, have developed a generative AI model that can screen millions of chemical compounds in days—a process that would take years using traditional methods. "We can—in a matter of days or hours—look at massive libraries of chemical compounds to identify those that display antibacterial activity," explains Collins, professor of medical engineering and science at MIT. Collins and his team trained their AI model by feeding it the chemical structures of known antibiotics, allowing the algorithm to learn what makes a compound effective at killing bacteria. They then unleashed the AI on a library of more than 45 million different chemical structures, searching for compounds that could target two of the most dangerous drug-resistant bacteria: Neisseria gonorrhoeae (which causes gonorrhoea) and Staphylococcus aureus (the bacteria behind MRSA infections). The results were striking. The AI designed 36 million potential compounds, and researchers selected 24 to synthesize in the laboratory. Seven showed some antimicrobial activity, and two proved highly effective at killing strains of both bacteria that were resistant to other antibiotics. Crucially, these compounds appear to attack the bacteria in entirely different ways than existing drugs, suggesting they could form a new class of medicines capable of overcoming bacterial defenses. Steps to Understanding AI-Driven Drug Discovery - Training Phase: Researchers feed AI models the chemical structures of known effective drugs so the algorithm learns what characteristics make a compound work against a specific disease. - Screening Phase: The trained AI rapidly analyzes millions of chemical compounds from digital libraries, scoring each one for its potential to target the disease in question. - Design Phase: AI uses generative techniques to build entirely new compounds from scratch, adding atoms and molecular bonds while continuously evaluating whether the emerging structure resembles a viable drug. - Validation Phase: Researchers synthesize the most promising AI-designed compounds in the laboratory and test them against actual pathogens to confirm effectiveness. Collins' laboratory has previously used this approach to discover other powerful antibiotic compounds, including drugs effective against Clostridium difficile (a common bowel infection) and Mycobacterium tuberculosis (which causes tuberculosis). Can AI Help Treat Parkinson's Disease? Beyond antibiotics, researchers are turning to AI to tackle diseases where even the basic biology remains mysterious. Parkinson's disease, first identified in 1817, still has no treatment that slows its progression more than two centuries later. More than 10 million people worldwide live with Parkinson's, and rates are rising in aging populations. In the United Kingdom alone, about 1 in 37 people will receive a Parkinson's diagnosis at some point in their lives. The challenge is that scientists still don't fully understand what causes Parkinson's. "There are endless debates about the origin of the disorder," says Michele Vendruscolo, professor in biophysics and co-director of the Centre for Misfolding Diseases at the University of Cambridge. "If you go to a Parkinson's conference, you will hear dozens of different hypotheses that are all actively investigated." This uncertainty has made it nearly impossible to design drugs that target the root cause. In 2024, Vendruscolo and his colleagues published research showing that machine learning could search for drug candidates capable of targeting the clumps of misfolded proteins that accumulate in Parkinson's patients' brains. These protein aggregations, called Lewy bodies, are thought to trigger the initial stages of neurodegeneration, eventually leading to tremors, slowness of movement, and muscle stiffness. Rather than trying to manage symptoms like current treatments do, Vendruscolo's team is using AI to identify compounds that could halt the disease's progression before symptoms appear. The current gold-standard treatment for Parkinson's is Levodopa, a drug that improves symptoms but can cause side effects such as involuntary movements. Vendruscolo's AI-driven approach represents a fundamentally different strategy: stopping the disease itself rather than treating its consequences. What Does This Mean for Future Medicine? These breakthroughs suggest that AI is opening a new era in drug discovery. By compressing years of research into days or hours, and by identifying entirely novel compounds that humans might never have conceived, artificial intelligence is tackling diseases that have resisted traditional approaches for decades. The two antibiotic candidates discovered by Collins' team are currently undergoing further testing, and similar AI-driven projects are targeting thousands of rare diseases with no known treatments. The implications extend beyond individual diseases. If AI can accelerate drug discovery for antibiotics and neurodegenerative diseases, the same approach could be applied to cancer, heart disease, and countless other conditions. The bottleneck in modern medicine may no longer be our understanding of disease biology, but rather our ability to design drugs that address it—and AI appears uniquely suited to solving that problem.
