Document resource
Background Over 40% of patients with inflammatory bowel disease (IBD) fail to respond to biologics, and no reliable biomarkers are currently available to predict treatment response before therapy initiation. Therefore, we aimed to evaluate the potential of gut microbial signatures as predictive biomarkers of biologics response in IBD and develop a noninvasive predictive tool.Methods Through a systematic literature review, we analyzed baseline fecal metagenomic data from public cohorts treated with biologics. Comparative profiling was conducted between the response and non-response groups. Microbial signatures associated with treatment outcomes were identified to construct a machine learning model, which was validated in an independent cohort.Results Metagenomic data from 175 IBD patients across five public datasets and 52 patients from an in-house cohort were processed. No significant differences were observed in alpha diversity between responders and non-responders, whereas beta diversity differed significantly ( IDDF2026-ABS-0208 Figure 1(A-E) Multi-cohort comparison of gut microbiome composition, functional differences, and predictive modeling of biologic response in IBD). After adjusting for confounding factors, 10 species and 3 pathways were enriched in responders, while 6 species and 3 pathways were more abundant in non-responders (IDDF2026-ABS-0208 Figure 1(F-G) Multi-cohort comparison of gut microbiome composition, functional differences, and predictive modeling of biologic response in IBD). Pathways enriched in responders were positively associated with responder-enriched taxa, whereas pathways enriched in non-responders showed positive associations with non-responder-enriched taxa (IDDF2026-ABS-0208 Figure 1(H) Multi-cohort comparison of gut microbiome composition, functional differences, and predictive modeling of biologic response in IBD). A total of 17 features (14 bacterial species and 3 pathways) were selected for model construction. The CatBoost model achieved an area under the curve (AUC) of 0.74 for predicting non-response, with a sensitivity of 82.4% and specificity of 66.7% in the test dataset. External validation yielded an AUC of 0.78, with balanced sensitivity (69.2%) and specificity (71.8%) (IDDF2026-ABS-0208 Figure 1(I) Multi-cohort comparison of gut microbiome composition, functional differences, and predictive modeling of biologic response in IBD).Conclusions Gut microbial taxonomic and functional features can serve as reliable predictive markers. The noninvasive microbiome-based model demonstrated robust performance in predicting treatment non-response, supporting the potential to guide personalized therapy.Abstract IDDF2026-ABS-0208 Figure 1