P1322 IBD, IBS, and Functional Dyspepsia: Differentiating The Spectrum via Microbiome based Machine Learning Approach

Abstract Background Alterations in gut microbial composition are linked to both inflammatory and functional gastrointestinal (GI) disorders. Although inflammatory bowel disease (IBD) and irritable bowel syndrome (IBS) share overlapping symptoms, their distinct microbial signatures—along with those of functional dyspepsia (FD)—remain incompletely defined. This study compared gut microbiome profiles across healthy controls (HC), IBS, IBD, and FD within a large, geographically diverse Indian cohort. Methods 487 consecutive patients with 1 or more GI symptoms (3m duration):abnormal bowel movements,epigastric pain,weight loss, blood/mucus in stools,bloating were prospectively enrolled from Jul’24-Sept’25. Individuals with GI surgery/recent probiotic or antibiotic use were excluded. 90 asymptomatic individuals were enrolled as healthy control. Fecal samples were collected pre-diagnosis. Upon diagnostic tests as per Rome IV criteria and colonoscopy, they were divided as IBD, IBS (IBS-Constipation/IBS-Mixed/IBS-Diarrhoea) and FD. Fecal DNA was subjected to 16S-V4 sequencing followed by ASV–based taxonomic classification and analysed using Jamovi. A machine learning approach using R-based scikit learn (XGBoost) was applied for pairwise disease classification and IBS subtyping. Model performance used 70:30 train–test split and AUC metrics. Results After diagnostic confirmation, 522 individuals were included; HC = 90,IBS = 194,FD = 180,IBD = 58 (68% males;median age 47.5y IQR: 36–57. Those with malignancy or involving surgery were excluded. Pro-inflammatory Proteobacteria abundance progressively increased from HC → FD → IBS → IBD (R² = 0.9), while beneficial Actinobacteria showed an inverse trend (R² = 0.87)(Fig1A). At the genus level, Streptococcus enrichment and reduction in Ruminococcus, Roseburia, and Ligilactobacillus followed similar gradients (R² = 0.9)(Fig1B). The Firmicutes:Bacteroides ratio declined sharply across all disease groups vs HC (p 0.001) but did not differ markedly among them. Group-wise and pairwise comparisons identified differentiating taxa (Fig2A). ML-Model achieved 80% AUC for differentiating IBD from non-IBD. Accuracy declined (60%) when distinguishing IBS vs FD (Fig2B). Key discriminatory taxa based on ML-based algorithm were identified (Fig2). IBS subtyping reached 76% accuracy. Among subtypes, IBS-C resembled HC, whereas IBS-D aligned more closely with IBD, characterized by increase in Proteobacteria & decrease in Roseburia (Fig1C). Conclusion This study reveals a continuum of gut microbial alterations spanning HC, FD, IBS, and IBD, supporting the concept of a microbiome spectrum from non-inflammatory to inflammatory disease states. ML–based models demonstrate the diagnostic utility of microbiome profiling, for future personalized therapies. Conflict of interest: Thakur, Manisha: No conflict of interest Raghunathan, Nalini: None Gunala, Nikhil: No conflict of interest Patel, Rajendra: No conflict of interest Mekala, Dhanush: No conflict of interest Dr. Banerjee, Rupa: RB has received grants/research support from Asian Healthcare Foundation, and the Leona M and Harry B Helmsley Charitable Trust Advisory board fees from Abbott, AstraZeneca, Abbvie, Cadila, Cipla, Dr Reddy Labs, Eli Lilly, Emcure, Ferring Pharma, Hetero Drugs, Janssen, MSN Labs, Mankind Pharma, Menarini, Micro Labs, Pfizer, Sun Pharmaceuticals, Takeda Pharmaceuticals, Torrent, Waterley, and Zydus.