Nearly a quarter of non-antibiotic drugs can kill bacteria at prescribed doses
Since Alexander Fleming's discovery upended the course of human suffering in 1928, antibiotics have been both salvation and slow-burning crisis — their overuse quietly teaching bacteria to survive them. Now, researchers at UMass Chan Medical School have turned to machine learning to ask an old question in a new way: could drugs designed for cancer, diabetes, or depression hold the keys to killing bacteria through mechanisms medicine has never thought to use? By mapping nearly two million drug-bacteria interactions, they have found that a quarter of non-antibiotic drugs do indeed kill bacteria — and do so through pathways that existing antibiotics have never touched, offering a rare opening in a fight the world is slowly losing.
- The World Health Organization recorded 1.27 million deaths from antibiotic-resistant bacteria in 2019 alone, and the pace of resistance is outrunning the pace of discovery.
- A surprising quarter of drugs prescribed for chronic conditions like cancer and diabetes can kill bacteria at normal doses — but without knowing how, their potential has remained locked away.
- Researchers built a machine learning system that analyzed nearly two million toxicity interactions between 200 drugs and thousands of mutant bacterial strains, mapping the hidden logic of how each drug kills.
- Non-antibiotic drugs clustered entirely apart from traditional antibiotics on the resulting network map, signaling they exploit bacterial vulnerabilities that standard treatments have never targeted.
- The team confirmed one such novel target — a bacterial protein hit by the parasite drug triclabendazole — proving the method can pinpoint mechanisms even before researchers fully understand them.
- The approach now offers a practical filter for drug discovery, steering researchers toward genuinely novel compounds and breaking a long-standing bottleneck in the search for new antibiotics.
Antibiotics reshaped medicine after 1928, turning once-fatal infections into manageable conditions — but their overuse set a slower catastrophe in motion. Bacteria evolved defenses, and by 2019 the WHO was counting 1.27 million deaths a year from resistant strains. The pipeline for new antibiotics has struggled to keep pace.
Researchers at UMass Chan Medical School pursued an unexpected lead: nearly a quarter of drugs designed for entirely different purposes — cancer, diabetes, depression — can kill bacteria at doses already considered safe. The critical question was whether they did so through mechanisms distinct from traditional antibiotics. If they did, they could serve as blueprints for a new class of treatments. If they merely echoed existing antibiotics, their widespread use in chronic disease care might quietly accelerate resistance.
To find out, the team built a machine learning method that processed close to two million interactions between 200 drugs and thousands of mutant bacterial strains, tracking which genes and cellular processes shifted as bacteria responded to each compound. When the results were mapped, known antibiotics sorted themselves into logical clusters — cell-wall disruptors in one group, DNA-replication blockers in another. Non-antibiotic drugs formed entirely separate hubs, suggesting they were reaching bacteria through pathways standard antibiotics had never exploited.
The confirmation came through evolutionary experiments: researcher Carmen Li grew bacteria through hundreds of generations of exposure to non-antibiotic drugs and sequenced the mutations that emerged. This allowed the team to identify the precise bacterial protein targeted by triclabendazole, a parasite medication — a target current antibiotics typically ignore. Two other non-antibiotic drugs operating through similar mechanisms hit the same mark, validating that the machine learning maps could match drugs by killing strategy even before their mechanisms were fully understood.
The broader significance lies in what the method changes about the search itself. Drug discovery has long been a filtering problem — thousands of candidates screened, most discarded because they work like antibiotics already in use. This approach acts as an early guide, surfacing the rare compounds that attack bacteria in genuinely novel ways and pointing toward bacterial targets medicine has not yet exhausted.
Antibiotics transformed medicine in 1928, turning deadly infections like pneumonia and tuberculosis into treatable conditions and making surgery safer than it had ever been. But that triumph came with a hidden cost. When antibiotics are overused, bacteria evolve defenses against them. The World Health Organization documented 1.27 million deaths from antibiotic-resistant bacteria in 2019 alone, and the threat is accelerating.
Researchers at UMass Chan Medical School have found an unexpected path forward: nearly a quarter of drugs never designed as antibiotics—medications for cancer, diabetes, depression—can kill bacteria at the doses doctors already prescribe them. The puzzle was figuring out how. If these non-antibiotic drugs attacked bacteria using entirely different mechanisms than traditional antibiotics, they could become blueprints for a new generation of treatments. But if they worked the same way, their long-term use in chronic disease treatment might actually speed up resistance.
To solve this, the research team developed a machine learning method that analyzed almost two million instances of toxicity between 200 drugs and thousands of mutant bacteria strains. The approach worked by identifying which genes and cellular processes changed when bacteria mutated in response to different drugs. By mapping how these changes affected bacterial survival, the researchers could infer the killing mechanisms each drug employed.
The results were striking. When the team plotted known antibiotics on their network map, drugs with similar killing mechanisms clustered together naturally. Antibiotics that attacked the bacterial cell wall grouped separately from those that disrupted DNA replication—exactly as expected. But when non-antibiotic drugs were added to the analysis, they formed their own distinct hubs, completely separate from the antibiotic clusters. This suggested that non-antibiotics were hitting bacteria through pathways that standard antibiotics had never exploited.
The breakthrough came when colleague Carmen Li sequenced bacteria that had evolved resistance to various non-antibiotic drugs over hundreds of generations. By identifying which genes had mutated, the team pinpointed the exact bacterial protein targeted by triclabendazole, a drug normally used to treat parasite infections. Crucially, current antibiotics don't typically target this protein. Two other non-antibiotic drugs that worked through similar mechanisms also hit the same target, proving that the machine learning maps could identify drugs with matching killing strategies even before their mechanisms were understood.
The implications ripple outward. Drug discovery has always been a numbers game—researchers screen thousands of chemicals, most of which work like existing antibiotics and get discarded. This new method acts as a filter, identifying the rare compounds that attack bacteria in genuinely novel ways. It addresses what has long been a critical bottleneck in antibiotic development: finding not just any drug that kills bacteria, but one that works through a mechanism medicine hasn't already exhausted. The research suggests there are still unexploited pathways to bacterial death, still roads available in the fight against resistant infections.
Citações Notáveis
There are different ways to kill bacteria we haven't exploited yet, and there are still roads we can take to fight the threat of bacterial infections and antibiotic resistance.— UMass Chan Medical School researcher
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that non-antibiotics kill bacteria differently than antibiotics do?
Because if they use the same mechanisms, long-term use could breed resistance to those pathways too. But if they're different, they're potential blueprints for entirely new classes of antibiotics.
So you're saying we've been missing an obvious resource?
Exactly. Thousands of drugs already approved for human use have been sitting there with antibacterial properties we didn't fully understand. We were looking for new antibiotics when we already had candidates in the medicine cabinet.
How does the machine learning actually work?
It watches what happens when bacteria mutate in response to different drugs. Drugs that cause similar mutation patterns are grouped together—they're likely killing bacteria the same way. The non-antibiotics formed separate clusters, which told us they were doing something different.
And that's useful because?
Because drug discovery is expensive and slow. Most candidates fail because they work like antibiotics we already know about. This method lets us spot the rare ones that don't, the ones that might actually matter.
What happens next?
We need to understand exactly how these non-antibiotics work at the molecular level. And we need to test whether using them long-term in chronic disease treatment actually prevents resistance, or whether we'd face the same problem we have now.
Is this a solution to antibiotic resistance?
It's a tool. A good one. But resistance will keep evolving. We need multiple strategies—better stewardship, new mechanisms, maybe combination therapies. This opens one door that was locked before.