AI in IBD research
Increasing antimicrobial resistance is an emerging cause of harm and death, with an estimated 4.7 million deaths worldwide associated with it. Antibiotic-resistant bacteria are more common in patients with inflammatory bowel disease (IBD), and even antibiotics previously used to treat resistant strains are losing efficacy due to the development of further resistance. These antibiotics also affect the entire gut microbiome by killing beneficial bacteria involved in vitamin synthesis and in preventing harmful resistant bacteria from proliferating. This is particularly concerning for patients with IBD, who already have a dysregulated gut microbiome.
A 2025 study published in Nature Microbiology was the first in the world to utilise artificial intelligence (AI) to predict how an antibiotic would function. Through this approach, researchers discovered a new antibiotic that may reduce the adverse effects of antibiotic use in patients with IBD. Over 10,000 molecules were screened, with only those showing some antibacterial activity selected. One of the promising candidates, named ‘enterolin herin’, demonstrated high efficacy against certain resistant bacteria while sparing the non-harmful bacteria normally present in the gut.
An AI deep learning program called DIFFDock* was then used to help identify the mechanism of action of this potential new antibiotic. Remarkably, the predictions it generated aligned with experimental findings. DIFFDock achieved this by modelling how molecules interact with one another.
While the discovery of this new antibiotic is promising, its effects in the human body remain unknown, and it is unlikely to be available to IBD patients in the near future. Further studies are needed to first establish its safety in humans and then confirm its efficacy. The use of AI to predict how medications will work is a novel approach in research, but it shows great promise in accelerating the discovery of new treatments.
*DIFFDock is an AI-powered molecular docking tool that predicts how small molecules (like drugs) bind to proteins using diffusion models. It represents a major advance in computational drug discovery because it treats docking as a generative modelling problem rather than a traditional optimisation task