The bacteria that causes tuberculosis, Mycobacterium tuberculosis

The bacteria that causes tuberculosis, Mycobacterium tuberculosis Photo: Kateryna Kon / Shutterstock

Tufts Scientists Use Artificial Intelligence to Improve Tuberculosis Treatments

Researchers in the Aldridge Lab have devised rules for a faster, more effective way to identify potential new drug cocktails against this infectious disease

Imagine you have 20 new compounds that have shown some effectiveness in treating a disease like tuberculosis (TB), which affects 10 million people worldwide and kills 1.5 million each year. You know that to treat the disease effectively, patients will need a combination of three or four drugs because TB bacteria behave differently in different environments—and in some cases, evolve to become drug-resistant.

How do you decide which drugs to test together? Twenty compounds in three- and four-drug combinations offer nearly 6,000 possible combinations.

In a recent study, published in the September issue of Cell Reports Medicine, researchers from Tufts University used machine learning to design a data-driven solution to this challenge that will allow researchers to consider novel drug combinations at a new scale. They believe their new system can increase the speed at which scientists determine which drug combinations will most effectively treat tuberculosis, the second leading infectious killer in the world.

“Our framework creates accurate predictions of how effective treatments will be when we move from testing in a lab to testing in mouse models, which is an important step in choosing which treatments progress to human clinical trials,” says Bree Aldridge, associate professor of molecular biology and microbiology at Tufts University School of Medicine and of biomedical engineering at the School of Engineering, and a member of the immunology and molecular microbiology program faculty at the Graduate School of Biomedical Sciences. “It provides a more efficient avenue for determining what multi-drug treatments should be moved to preclinical studies. And it provides interpretable rules to help researchers and drug developers construct their own optimal combinations of drugs to test.”

“We now have empirical evidence of how drug combinations should be designed, rather than guessing and testing in animals for every combination,” adds Jonah Larkins-Ford, GBS22, co-first author on the paper and a principal scientist at MarvelBiome who completed his doctorate and postdoctoral research in the Aldridge Lab. “The concrete rules we’ve developed provide a path for deliberate, data-driven design to identify new three- and four-drug cocktails to better treat tuberculosis.”

Aldridge’s lab developed and uses DiaMOND, or diagonal measurement on-way drug interactions, a method to systemically study pairwise and high-order drug combination interactions to identify shorter, more efficient treatment regimens for TB and diseases that require combination drug therapy, like cancer. With the design rules established in the new study, the researchers believe their new system can increase the speed at which scientists determine which drug combinations will most effectively treat tuberculosis.

The Pain of TB Treatments

Until COVID-19 appeared in 2020, TB was the deadliest infectious disease on the planet. Its treatment is a cocktail of four drugs taken for four months or more with significant side effects that make it difficult for many patients to complete the treatment regimen. TB is also becoming increasingly resistant to standard treatments.

“The drugs used to treat TB make patients feel sick and can result in weeks of lost work time and productivity,” says Aldridge, who is corresponding author for the paper and also associate director of Tufts Stuart B. Levy Center for Integrated Management of Antimicrobial Resistance. “As a result, staying the course is hard.”

Bree Aldridge is also associate director of the Tufts Institute for Artificial Intelligence.

Bree Aldridge is also associate director of the Tufts Institute for Artificial Intelligence. Photo: Alonso Nichols / Tufts University

Tuberculosis infects the lungs and is spread person to person through tiny droplets released into the air when someone coughs or sneezes. It is most common in Latin America, Africa, Asia, and Eastern Europe. It is also more common in individuals with HIV/AIDS or weakened immune systems. A person can be infected with TB but have no symptoms (latent TB) for weeks, months or even years.

Promising new antibiotics aimed at TB have been developed recently. But researchers and drug developers have been challenged to find ways to cheaply and effectively determine which drugs will work best in combination. It’s a mathematically complex, expensive, and time-consuming process using traditional procedures, Aldridge says.

During drug development, scientists first test potential pharmaceuticals in the lab using a plate with slots containing samples of TB bacteria. Treatments that kill TB on the plate proceed to testing in lab rodents, and those regimens that are effective in rodents proceed to clinical trials in humans.

Combination therapy is necessary because the TB bacterium Mycobacterium tuberculosis is highly adaptive to its site of infection, which vary even within one patient, giving rise to differences in drug response among the bacteria in one person. Also, individuals may be infected with a drug-resistant strain, or their TB infection may evolve over time to become resistant to one or more of the traditional cocktail’s antibiotics.

“It’s a bit like putting together a chicken Caesar salad,” Aldridge explains. “Each drug in the existing four-drug cocktail performs a certain function to kill the bacteria, just like the four major elements of a chicken Caesar salad—lettuce, dressing, croutons and chicken—combine to provide the texture, flavor, crunch, and protein that yield a tasty, sustaining meal.”

When considering new drug combinations, the salad analogy applies: will a particular drug pair under consideration be the equivalent of the crouton, lettuce, the chicken, or the dressing? And if it is the crouton, is it the tastiest, crunchiest one available?

 An Assist from Artificial Intelligence

For the study, the researchers used data from large studies that contained laboratory measurements of two- and three-drug combinations of twelve commonly used anti-tuberculosis drugs. Using mathematical models and artificial intelligence, the Tufts team discovered a set of rules that drug pairs need to satisfy to be potentially good treatments.

“A drug pair doesn’t need to do everything we need,” says Aldridge, who is also associate director of the Tufts Institute for Artificial Intelligence. “They just need to do one thing and do it well, whether being able to kill one type of TB bacteria or do it under particular conditions.”

This new modeling system and the use of drug pairs (rather than combinations of three or four drugs) cuts down significantly on the amount of testing that needs to be done before moving a drug pair into further study.

“Using the rules we’ve established and tested, we can substitute one drug pair for another drug pair as long as we think it solves the same problem—i.e. we can substitute one type of crouton for another—and know with a high degree of confidence that the drug pair should work in concert with the other drug pair to kill the TB bacteria in the rodent model,” says Aldridge.

With multi-drug resistant TB on the rise, and the ever-present risk of people with TB not taking the drugs consistently for a long period of time, it is increasingly important to find new cocktail combinations that work as well as or better than the existing mix, and to find treatment combinations that work more quickly than four to six months.

“This system we developed should make the process of achieving that goal faster and cheaper,” she says.

Citation: Research reported in this article was supported by the Bill & Melinda Gates Foundation under award number OPP1189457 and by the National Institutes of Health under award number 1U54CA225088. Complete information on authors, funders, and conflicts of interest is available in the published paper.

DISCLAIMER: The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Back to Top