Every year, tuberculosis kills more people than any other infectious disease on Earth. Now, a team of scientists in Texas is using artificial intelligence to hunt for new drugs against this ancient enemy — and their tools are already showing remarkable accuracy.
When researchers test thousands of compounds to find ones that might fight tuberculosis, they often end up overwhelmed. "We might get thousands of compounds from a screen and then have to decide which one are we going to work on?" said James Sacchettini, Ph.D., a professor at Texas A&M University. Many look promising at first, only to prove costly dead ends years later.
Sacchettini's laboratory has built AI tools to help scientists focus their efforts on the compounds most likely to succeed. One of those tools, called CAGE-Fusion, can sort through early test results and flag compounds that only appear to work. These "nuisance molecules" might clump together, trick the test's chemical signal, react in unwanted ways, or stick to too many targets instead of just one.
The results are striking. Given one misleading compound and one clean one, the model ranks the problematic one as more suspicious about 94 percent of the time. The AI is especially good at catching compounds that react instead of binding, though it struggles more with compounds that stick to too many targets. Siddhant Rath, an AgriLife Research scientist who led the model's development, said chemists can see exactly which parts of a molecule raised red flags — not just a yes-or-no answer.
The work matters because tuberculosis drug discovery is painfully slow. The bacteria hide behind a thick, waxy coating that keeps most drugs from reaching their targets. They also grow slowly, so a single experiment can take months, whereas similar tests with staph or strep take just a week. "That's part of the reason why the drug discovery pipeline has been relatively slow," Sacchettini said. "It's a perfect area to work with AI."
Tuberculosis has plagued humans for thousands of years. Standard treatment takes months, and drug-resistant strains or cases involving HIV co-infection take even longer. Most of the estimated 10 million people who fall sick each year live in lower-income regions, where long treatment times and limited healthcare make the disease especially hard to control.
Before building their AI tools, Sacchettini's team tackled a simpler problem: research data in academic labs often sits scattered across hard drives, slide decks and researchers' memories. They created DAIKON, an open-source platform that tracks a drug target from gene to finished chemistry in one searchable place. The Gates Foundation supports the Tuberculosis Drug Accelerator partnership, which uses DAIKON across its network of labs and companies. The new AI tools plug directly into this system.
"If we can use AI to shorten the time it takes to go from an idea to a real treatment, that would be wonderful," Sacchettini said. The CAGE-Fusion model runs automatically inside DAIKON, flagging likely problems before a compound reaches further, expensive stages of testing. For a disease that has outpaced medicine for centuries, even a small dent in the timeline could save countless lives.