Lab Robots and AI Identify Bacterial Trio That Boosts Gut Health

A biomedical engineering team at Duke University has developed a method to systematically search trillions of possible combinations of gut bacteria and dietary fibers, instead of testing them by trial and error. The results, published on July 27, 2026 in the journal Nature Chemical Biology, identify a specific trio of bacteria and one type of fiber that, combined, reliably produce butyrate, a compound crucial for gut health, regardless of what other microorganisms are already living there.

A method instead of guesswork

Study lead Ophelia Venturelli, associate professor of biomedical engineering at Duke University, explains that the prevailing approach to designing probiotic and prebiotic supplements has relied mostly on intuition and limited testing. Companies paired bacteria with fibers hoping for a synergistic effect, but without a systematic way to check which combinations actually work amid the variability of the human microbiome.

The problem is that gut microbiome composition varies drastically between people, depending on diet, medications taken, and organisms already present in the gut. A combination of bacteria and fiber that works for one person can fail completely in another. The Duke team wanted to find combinations that produce a repeatable effect regardless of that background.

Robots and active learning

The key to searching such a vast space of possibilities was pairing lab automation with machine learning algorithms based on Bayesian optimization. Lab robots ran thousands of parallel bacterial culture experiments in different mixtures, while an AI model analyzed the results of each round and used them to propose which combinations to test in the next round, gradually narrowing the search.

Across five rounds of high-throughput tests, each simultaneously examining up to 390 conditions, the team was able to explore a fraction of the trillions of theoretically possible combinations, but enough to land on one that produced a consistent, predictable result.

Probiotics on their own may not colonize the gut long enough - Ophelia Venturelli, associate professor of biomedical engineering, Duke University

What they found

The identified trio, the bacteria Bacteroides uniformis, Anaerostipes caccae and Prevotella copri combined with inulin fiber, consistently produced butyrate, a short-chain fatty acid that nourishes the cells lining the colon, supports the gut barrier, and has documented anti-inflammatory effects. According to the team, the effect held regardless of what other microorganisms were already present in the tested environment, which is crucial, since that very variability has so far undermined the reproducibility of commercial products' results.

Implications for industry and medicine

The global probiotics and prebiotics market, worth about $130 billion, has struggled for years with inconsistent effects: a product that works in clinical trials for one group of patients produces no result in another. The method developed at Duke offers a way to design combinations with guaranteed reproducibility, which could lead to more targeted products tailored to specific gastrointestinal conditions.

The study's authors also point to broader clinical applications, including potential support for treating inflammatory bowel disease and other disorders in which an imbalanced microbiome plays a role. The researchers stress that the next step is testing the identified combination in animal models and, eventually, in human trials.

Wider context

The study fits into a growing trend of using lab robotics paired with AI to speed up biomedical discovery, from drug design to microbiome engineering. Automation enables a scale of experimentation impossible to achieve by hand, while active learning algorithms minimize the number of trials needed, steering researchers straight toward the most promising combinations instead of searching at random.

The project was funded by the US National Institutes of Health and the Army Research Office. Alongside Venturelli, the team included Bryce M. Connors, Jaron Thompson, Manasi Subhash Gangan, Nick Quinn-Bohmann, Sean M. Gibbons, and researchers from other institutions collaborating on the project.