According to a press release from Texas A &M, researchers are using artificial intelligence to narrow drug choices and make shared data easier to use.
Tuberculosis drug discovery speeds up with new AI tools at Texas A&M AgriLife
When scientists search for new drugs to fight tuberculosis, they often start with a long list of chemicals. Many look promising at first, but later they turn out to be poor choices—costing time and resources before teams can focus on the best options.
“Sometimes you may get thousands of compounds from an initial screening and then have to decide which ones are worth pursuing,” said James Sacchettini, Ph.D., a Texas A&M AgriLife Research scientist and professor. “Our goal is to use AI to help researchers make better decisions sooner.”
Sacchettini’s lab has developed new AI tools to help with two major challenges in tuberculosis drug discovery: cutting down the time it takes to choose promising candidates after early tests, and organizing large amounts of research data so scientists can quickly find useful information.
Tackling a disease that takes too long to treat
Tuberculosis remains one of the world’s most serious infectious diseases. Treatment can take months, and longer timelines can be even more difficult for people facing drug-resistant forms of the disease or living with other health conditions such as HIV.
Part of what makes TB harder to study is biology itself. The bacterium has a tough, waxy outer layer that makes it difficult for many drugs to reach their targets inside the cells. The bacteria also grow slowly, meaning lab tests often take far longer than similar experiments for other infections.
“Tuberculosis experiments can take months,” Sacchettini said. “That slow pace is one reason AI is a good fit for this area—it can help reduce delays and focus efforts.”
Building on a shared data foundation
Before these new AI systems, Sacchettini’s team worked on a simpler need: making sure research results are connected and searchable. In many labs, information can be scattered across files, shared drives, and individual researchers’ notes.
To address this, his team created DAIKON, an open-source platform that helps track a TB drug target from early information through years of chemistry work. The platform was published in 2023, and the Gates Foundation-supported Tuberculosis Drug Accelerator (TBDA) uses DAIKON across its network. New AI tools from Sacchettini’s lab connect directly to this system.
The approach is not meant to replace scientists’ judgment. Instead, AI helps teams save time by helping identify what is unlikely to work, so researchers can focus on stronger leads.
Flagging false signals earlier in screening
In the early stages of drug development, researchers test many compounds against a protein to see whether any appear to interfere with the target. But some compounds produce “false positives”—signals that may be caused by issues with the compound itself rather than real drug activity.
Sacchettini’s lab developed an AI model called CAGE-Fusion to recognize these misleading results. Using published screening data, the model sorts compounds into four categories of trouble, including compounds that clump together, compounds that interfere with the test’s signal, compounds that react instead of binding, and compounds that attach to many targets rather than just one.
According to the study that described the model, when comparing a known nuisance compound to a clean compound, the model ranks the nuisance compound as more suspicious about 94% of the time. The system can also flag likely problems automatically inside DAIKON, before a compound moves into later and more expensive testing stages.
Chemists can also see why the model flagged a compound, including which parts of the molecule appeared problematic.
Turning consortium data into an easier-to-use resource
Beyond screening, TBDA involves many stages and many participating labs. Sacchettini’s team is also using AI to help researchers make better use of shared knowledge from across the pipeline.
Two researchers in the lab, Siddhant Rath and Saswati Panda, developed an AI system designed to help people quickly find and understand information stored across TBDA’s projects. The system brings together TBDA data, so researchers can trace a molecule’s story across different efforts, including where research ended without progress.
Researchers can also ask questions through a chat interface, which can return details such as who presented findings and provide access to related materials like slides.
New tools match today’s computing power
Rath and Panda noted that the availability of stronger computing power has made AI more practical in research than in the past. Together with Sacchettini, they emphasized that AI is most valuable when used to guide decisions—helping researchers avoid dead ends and spend more time on the most promising work.
“We’re not hoping for AI to give us the exact right answer,” Sacchettini said. “But it can help us figure out what not to work on, which helps direct us toward what we should be focusing on.”
