Gut Bacteria Loss Linked to Crohn’s Disease Scarring, AI Study Finds

If you have Crohn's disease and have watched a colonoscopy report shift from "inflammation" to "stricture" or "fibrosis" (permanent scar tissue that narrows the bowel), you already know that scarring is one of the complications doctors worry about most. A new study led by researchers at the University of Birmingham used generative AI to help analyze small, hard-to-study patient datasets, then combined that with real gut microbiome and gene-expression data to trace how a loss of certain gut bacteria tracks with the shift from ordinary inflammation to that harder-to-treat scarring.

KEY TAKEAWAYS

• Researchers found that beneficial, short-chain-fatty-acid (SCFA) producing gut bacteria were progressively depleted from healthy guts to inflamed Crohn's tissue to scarred (fibrotic) tissue, while two other bacterial groups, Bilophila and Bacteroides, became more abundant.
• The study used generative AI to create realistic synthetic gene-expression data to work around a small fibrotic patient cohort, then combined that with existing public datasets rather than collecting new patient samples.
• The microbiome and gene-expression data came from different groups of patients, not the same people, so this shows a correlation worth investigating further, not proof that these bacteria cause scarring.
• Predicting garden-variety Crohn's disease from gene activity worked well; predicting which patients had fibrosis specifically was harder and less reliable.

AI-Augmented Analysis Links Gut Bacteria Loss to Crohn's Disease Scarring

Researchers found that as gut tissue moves from healthy to inflamed Crohn's disease to fibrotic (scarred) Crohn's disease, several genera of short-chain-fatty-acid (SCFA) producing bacteria, byproducts that feed and calm the gut lining, become steadily scarcer, according to the study published in Frontiers in Artificial Intelligence. The depleted group included Faecalibacterium, Anaerostipes, Coprococcus, and Ruminococcus, all bacteria known for producing butyrate and other SCFAs that help keep the gut lining calm and intact. At the same time, two other genera, Bilophila and Bacteroides, became more abundant, with the increase most pronounced in the fibrotic tissue samples. The team, led by corresponding author Animesh Acharjee and colleagues at the University of Birmingham and King Faisal Specialist Hospital in Jeddah, arrived at this picture by combining public RNA sequencing data, single-cell sequencing data, and 16S rRNA gut bacteria profiling (a common way to identify which bacterial genera are present in a stool or tissue sample) rather than running a new prospective trial.

Why Crohn's Disease Patients Facing Strictures Should Care

Intestinal fibrosis, the buildup of scar tissue in the bowel wall, is one of the complications that turns Crohn's disease from a manageable condition into one that sometimes requires surgery to remove a narrowed (stricture) section of bowel. Unlike active inflammation, which anti-inflammatory drugs can often calm down, fibrotic scar tissue does not reverse with current anti-inflammatory therapy, which is part of why researchers are hunting for early markers, whether in blood, stool, or gut bacteria, that flag who is heading toward a stricture before it happens. If gut bacteria composition really does shift ahead of or alongside that transition, it opens a door to catching the process earlier, though this study is a first, exploratory step in that direction rather than a ready-to-use test.

How Generative AI Filled a Gap in a Small Fibrotic Patient Cohort

The researchers drew on existing public datasets: bulk gene-expression profiles from 176 people without inflammatory bowel disease (IBD), 187 with baseline (non-fibrotic) Crohn's disease, and 85 with fibrotic Crohn's disease, plus single-cell sequencing data and 16S rRNA gut bacteria profiles from a separate group of 80 samples (31 non-IBD, 38 baseline Crohn's, and 11 fibrotic Crohn's) after quality filtering. Because the fibrotic transcriptomic group was comparatively small, the team built a hybrid tool combining a generative language model (OpenAI's GPT-4.1-mini) with a statistical gene-expression simulator called SPsimSeq to generate synthetic patient samples that mimicked the statistical patterns of real fibrotic gene expression, guided by a curated list of 40 genes already tied to fibrosis in the published literature. They validated that the synthetic data preserved realistic gene-expression patterns using statistical comparison methods, then used both real and AI-augmented data to train machine-learning models. Those models classified baseline Crohn's disease from non-IBD tissue quite accurately (area under the curve, or AUC, a 0 to 1 accuracy score, of roughly 0.95 to 0.96), but classifying fibrotic tissue specifically was harder, with AUCs closer to 0.65 to 0.69 on real data alone and up to 0.82 once the AI-augmented data was added in, though that improvement came with a wide margin of uncertainty.

What This Could Mean If You're Managing Crohn's Disease

For someone living with Crohn's disease, this study is not a signal to start testing your microbiome for these specific bacteria or to reach for a supplement aimed at boosting SCFA producers. No probiotic, prebiotic, or diet change has been tested here, and the researchers did not establish that restoring these bacteria prevents or reverses fibrosis. What it does add is another data point worth raising with a gastroenterologist if you are being monitored for stricturing disease: researchers are increasingly looking at gut bacteria patterns, alongside genes like TGF-beta and TNF-alpha (both involved in inflammation and tissue repair) that the AI-augmented models flagged as informative, as part of a broader push toward earlier, less invasive ways to flag who is at risk of scarring.

Key Caveats: Small Cohorts and Two Separate Patient Groups

This is a secondary analysis of existing public datasets, not a prospective clinical trial that followed patients over time, and the authors are explicit that most of the underlying data was collected at a single point in time, which limits any ability to say what changed first. The fibrotic subgroups were also small (85 people in the gene-expression data, just 11 in the microbiome data), which the authors acknowledge reduces how well the findings might generalize. Perhaps most importantly, the gene-expression data and the microbiome data came from different groups of patients rather than matched samples from the same people, so the study could not directly link one person's bacterial profile to their own gene activity; the connection between the two datasets is inferential, not measured in the same individuals. The authors also note they did not run functional experiments to confirm that the bacterial changes actually reduce SCFA production in these patients, and the synthetic data, while statistically validated, has not yet been tested against an independent, real-world fibrotic cohort.

What Comes Next for Microbiome-Based Fibrosis Prediction

The authors frame this as a proof-of-concept for using generative AI to work around small, hard-to-collect patient cohorts in fibrosis research, not a finished diagnostic tool. The logical next steps are studies that collect matched gut bacteria and gene-expression samples from the same patients over time, and validation of both the bacterial signature and the AI-augmented models in independent Crohn's disease cohorts before anything resembling a clinical test could be considered.

Common Questions

What is intestinal fibrosis in Crohn's disease? It is the buildup of scar tissue in the bowel wall as a result of long-term inflammation, which can narrow the intestine (a stricture) and sometimes requires surgery.

Which gut bacteria were linked to Crohn's disease scarring in this study? Faecalibacterium, Anaerostipes, Coprococcus, and Ruminococcus (all short-chain-fatty-acid producers) were depleted, while Bilophila and Bacteroides were more abundant, especially in fibrotic tissue.

Was this a clinical trial in real Crohn's disease patients? No. It was a secondary, AI-augmented analysis of existing public datasets, including AI-generated synthetic data used to work around a small fibrotic patient group, not a prospective trial that followed patients over time.

Could this lead to a probiotic or supplement to prevent Crohn's disease scarring? Not based on this study alone. No intervention was tested, and the bacterial changes described here are an association observed in existing data, not a proven cause of fibrosis.

References

1. Daryll Philip, Daniela Santos, Sudip Mondal, Haneen Alomar, Georgios Gkoutos, Animesh Acharjee (2026). Generative AI-augmented transcriptomic and microbiome analysis across inflammatory and fibrotic disease states in Crohn’s disease. Frontiers in Artificial Intelligence. https://doi.org/10.3389/frai.2026.1881820
2. Fotios S. Fousekis, Konstantinos Mpakogiannis, Ioanna Nefeli Mastorogianni, Georgios D. Lianos, Dimitrios K. Christodoulou, Konstantinos H. Katsanos (2025). Intestinal Fibrosis in Crohn’s Disease: Pathophysiology, Diagnosis, and New Therapeutic Targets. Journal of Clinical Medicine. https://doi.org/10.3390/jcm14124060