A Multicenter Randomized Controlled Trial of Microbiome-Based Artificial Intelligence-Assisted Personalized Diet vs Low-Fermentable Oligosaccharides, Disaccharides, Monosaccharides, and Polyols Diet: A Novel Approach for the Management of Irritable Bowel Syndrome

Irritable bowel syndrome (IBS) is a prevalent functional gastrointestinal disorder characterized by recurrent abdominal pain, altered bowel habits, and bloating, with significant impacts on quality of life and economic burden. Pathophysiology involves gut-brain axis dysregulation, motility changes, visceral hypersensitivity, and gut microbiome dysbiosis. Emerging evidence implicates gut microbiome composition and function in IBS symptomatology, prompting interest in microbiome-targeted interventions. The low-FODMAP diet reduces fermentable carbohydrate intake and can alleviate symptoms, but long-term strict adherence may reduce beneficial microbial abundance and diversity. A microbiome-based personalized diet (PD), integrating microbiome profiling and AI algorithms, aims to tailor dietary recommendations to individual microbial profiles to enhance beneficial taxa and microbial diversity. This multicenter randomized controlled trial compares the efficacy and feasibility of an AI-assisted PD versus a FODMAP diet for IBS management, hypothesizing superior symptom improvement, patient satisfaction, and favorable microbiome changes with PD.

Design: Multicenter, parallel, randomized, controlled, double-blind comparative trial across four centers in three cities (Istanbul, Izmir, Kayseri), conducted August 2022–May 2023 per CONSORT guidelines; registered at ClinicalTrials.gov (NCT05646186). Participants: Adults aged 18–65 meeting Rome IV criteria for IBS; exclusions included severe chronic diseases, psychiatric comorbidity, other GI diseases, restrictive diets, colonoscopy or antibiotics within 4 weeks; probiotics/prebiotics/fecal assistance prohibited and monitored weekly. Randomization and masking: 1:1 allocation (blocks of 5) via computer-generated sequences sealed in opaque envelopes; research dietitians delivered interventions; investigators and data analysts were blinded to group allocation. Interventions: 6-week AI-assisted personalized diet (PD) vs low-FODMAP diet. PD was generated by Enbiosis personalized nutrition model using machine learning: baseline microbiome composition analyzed to compute an IBS index; Metropolis algorithm-based random walk perturbed compositions to identify lower-IBS-index nearby states; model estimated optimized micronutrient compositions and daily diets to modulate the microbiome toward healthier scores. Approximately 300 foods scored 0–10 for microbiome modulation: 0–3 avoid, 4–7 diversify, 8–10 needed; diets planned with 4–10 scored foods; raw greens and legumes restricted initially. FODMAP diet restricted fructans and galacto-oligosaccharides (e.g., wheat, rye, barley, onion, legumes), lactose, high free fructose foods (e.g., apples, pears, watermelon, asparagus, honey), and polyols (e.g., sorbitol, mannitol, maltitol, xylitol). Dietary overlap assessed via Jaccard index. Microbiome sampling and sequencing: Stool collected pre- and post-intervention; stored at −80°C. DNA extracted with Qiagen Power Soil kit; quantified with Qubit dsDNA HS assay; 16S rRNA gene V4 region amplified (primers 515F/816R) and sequenced on Illumina MiSeq (250 bp paired-end; ≥50,000 reads/sample; 15% PhiX spike-in). Bioinformatics: QIIME2 pipeline with Trimmomatic (PHRED≥30), DADA2 ASV inference, taxonomy via Naive Bayes classifier trained on SILVA v132; alpha/beta diversities via scikit-bio; paired two-sided t-tests and Mann–Whitney U for independent comparisons; FDR via Benjamini–Hochberg; Bray–Curtis ordination via multidimensional scaling; computations with scikit-learn and SciPy. Outcomes and monitoring: Primary outcome was within-individual IBS-SSS change compared between groups. Secondary outcomes: IBS-QOL, HADS anxiety and depression, Bristol Stool Form Scale (BSFS) stool frequency/consistency, and microbiome alpha/beta diversity and taxa changes. IBS-SSS (0–500; mild 75–175, moderate 175–300, severe >300), IBS-QOL validated measure, HADS (0–21 per subscale). Patients kept daily stool diaries during the 6-week period. Adherence: Weekly self-rated compliance; compliant if “frequently” or “always” in ≥3 of 6 assessments. Statistics: SPSS v21; descriptive stats; paired t-tests/Wilcoxon signed-rank; significance P<0.05 (actual P values log-transformed for display); effect sizes via Cohen d; power based on d=0.8, 80% power, α=0.05, SD=100; subgroup analyses via χ²/independent t-tests; correlations via Pearson/Spearman; PERMANOVA for multivariate differences (99,999 permutations).

Participants: 149 randomized (PD n=75; FODMAP n=74); 28 withdrew; analyzed n=121 (PD n=70; FODMAP n=51). Baseline: No significant demographic or subtype differences; most common subtype IBS-C (PD 45.7%, FODMAP 46.4%). Primary outcome: Mean IBS-SSS change −112.7 (PD) vs −99.9 (FODMAP), P=0.29 (between-group). Within-group improvements (both groups): abdominal pain severity and frequency (P<0.001 each), abdominal distension severity (P<0.001), bowel habits dissatisfaction (P<0.001), life interference (PD P<0.001; FODMAP P=0.02), IBS-SSS total (P<0.001 each). IBS-QOL improved in both groups (P<0.001). HADS: anxiety improved (PD P<0.001; FODMAP P<0.001); depression improved (PD P=0.02; FODMAP P<0.001). BSFS: PD stool form score increased modestly (3.43±1.40 to 3.70±0.73; P=0.02); FODMAP change not significant (P=0.24). Subtypes: IBS-SSS—PD improved IBS-C (P<0.001), IBS-D (P=0.010; Cohen d=0.81), IBS-M (P<0.001); FODMAP improved IBS-C (P<0.001) and IBS-M (P<0.001), not IBS-D (P=0.312). IBS-QOL—PD improved IBS-C (P<0.001), IBS-D (P<0.01), IBS-M (P=0.08 borderline reported as 0.08); FODMAP improved IBS-C (P=0.04) and IBS-D (P=0.22 not significant), IBS-M (P=0.646 not significant). Diet overlap: Jaccard similarity index 0.29. Adherence: Most participants reported high compliance; PD slightly higher compliance than FODMAP (supplement). Microbiome: PD increased alpha diversity (Shannon) significantly, particularly in IBS-C and IBS-D; FODMAP showed no significant alpha diversity change. Baseline beta diversity did not differ between groups (IBS-C P=0.996; IBS-D P=0.457; IBS-M P=0.06); post-intervention, PD showed more pronounced beta-diversity shifts (PERMANOVA P<0.05 in IBS-C and IBS-M; IBS-D not significant). Taxa: PD increased Faecalibacterium prausnitzii (P<1e−4, paired t-test); decreased Ruminococcaceae (P<1e−5); no significant changes for these taxa in FODMAP (P>0.05). Completion rate 81%.

The trial addressed whether an AI-assisted microbiome-based personalized diet (PD) provides superior clinical and microbiome benefits over a standard low-FODMAP diet in IBS. Both interventions improved symptoms and quality of life, but PD achieved significant reductions in IBS-SSS and enhancements in IBS-QOL across all subtypes, with particularly notable effects in IBS-D, where FODMAP did not show significant improvement. PD also induced beneficial microbiome changes, including increased alpha and beta diversities and higher abundance of beneficial taxa such as Faecalibacterium prausnitzii, alongside reductions in Ruminococcaceae. These findings suggest PD can deliver comprehensive benefits—symptom relief, psychosocial improvements (anxiety, depression), and microbiome modulation—by tailoring diets to individual microbial profiles, potentially addressing heterogeneity in IBS pathophysiology better than a one-size-fits-all FODMAP approach. As a comparative trial using FODMAP as active comparator, the results underscore the value of personalized dietary strategies in IBS care.

This multicenter randomized controlled trial supports the effectiveness of an AI-assisted microbiome-based personalized diet in managing IBS symptoms and improving quality of life across subtypes, with additional favorable modulation of gut microbiome diversity and specific beneficial taxa. While the low-FODMAP diet remains a standard-of-care, integrating individualized dietary recommendations based on microbiome profiles may optimize outcomes. Future research should explore longer intervention durations, integration of genetic and metabolic profiling, identification of predictive biomarkers for dietary response, and long-term monitoring of microbiome and clinical effects to refine personalized nutrition strategies for IBS.

- Uneven group sizes post-dropout (PD n=70 vs FODMAP n=51), potentially affecting statistical power and between-group comparisons. - Attrition: 28 participants withdrew (overall completion rate 81%); differing drop-out rates between arms, though reasons were not systematically different; potential attrition bias mitigated by baseline comparability of completers. - No placebo or no-treatment arm by design; FODMAP served as active comparator. - Genetic and metabolic parameters of individuals were not assessed; an integrated approach may yield improved outcomes. - Intervention duration limited to 6 weeks; longer-term dietary modifications may produce different microbiome and clinical effects; need for studies comparing varying durations with long-term follow-up. - Generalizability may be influenced by exclusion criteria (e.g., restrictive diets, recent antibiotics/colonoscopy) and specific clinical setting.