INTEGRATING MULTI-OMICS AND ARTIFICIAL INTELLIGENCE FOR PRECISION MEDICINE IN PEDIATRIC GUT MICROBIOME DISORDERS: EMERGING BIOMARKERS, PREDICTIVE MODELS, AND PERSONALIZED THERAPEUTICS

Abstract

Background: Pediatric Inflammatory Bowel Disease (PIBD) presents significant diagnostic and therapeutic challenges in low-resource settings like Pakistan, where unique dietary and genetic factors influence disease pathology. This study evaluates the efficacy of integrating multi-omics data with Artificial Intelligence (AI) for precision medicine in a small-sample cohort from Islamabad. Methods: A prospective pilot study was conducted at PIMS, Islamabad (N = 40 discovery; N = 15validation), utilizing shotgun metagenomics, untargeted metabolomics, and host transcriptomics. To overcome sample size limitations, an XGBoost model optimized with SMOTE data augmentation and Leave-One-Out Cross-Validation (LOOCV) was employed. AI-guided personalized therapeutics were administered to PIBD patients for 12 weeks, targeting microbial restoration through locally adapted prebiotic and probiotic interventions. Results: The multi-omics XGBoost+SMOTE model achieved superior classification performance (92.5% accuracy, 94% AUC-ROC, 91% F1-score; p < 0.001), significantly outperforming clinical-only (65%) and single-omics models. Key predictive biomarkers included depleted Faecalibacterium prausnitzii (SHAP=0.89), reduced fecal butyrate (0.82), and downregulated host IL-10 mRNA (0.74). Baseline analysis revealed PIBD patients had significantly lower BMI Z-scores (-1.4 vs. +0.12; p < 0.05) and fiber intake (8.5 vs. 14.2 g/day; p < 0.001) compared to healthy controls. Post-intervention, patients demonstrated a 53% reduction in PCDAI scores, 62% decrease in fecal calprotectin, 211% increase in butyrate, and 462% enrichment of F. prausnitzii (all p < 0.001), alongside improved nutritional status. Conclusion: Integrating multi-omics with robust AI methodologies enables high-precision diagnostics and effective personalized therapeutics even in small pediatric cohorts. This approach successfully addresses local epidemiological factors in Islamabad, offering a scalable framework for precision medicine in resource-limited environments. Future work should focus on multi-center validation and integration of host genomics to further refine clinical applicability.