AI may enable earlier diagnosis and treatment pathways for Crohn's disease

What To Know

Researchers analysed hundreds of intestinal transcriptomic samples plus microbiome data and applied machine-learning, including generative models, to expand limited fibrosis datasets.

They identified a set of genes and three connected biological patterns (persistent immune activation, reduced epithelial/functional gene activity, and cellular stress/barrier weakening) that associate with fibrotic versus inflammatory tissue.

Microbiome changes — loss of short-chain–fatty-acid–producing bacteria and increases in taxa like Bilophila and Bacteroides — tracked with those gene-expression patterns. The study frames fibrosis not as a single final state but as an active, reinforcing process involving immunity, epithelial dysfunction, and microbial shifts.

The authors used AI-generated synthetic gene-expression profiles to bolster machine-learning training because fibrotic samples are relatively scarce; these synthetic samples were used to test model robustness, not to replace real patient data.