Network-Based Multiomic Nutrient-Associated Predictive Models for Inflammatory Bowel Disease.

Background: Inflammatory bowel disease (IBD) is a multifactorial disease involving a complex interplay between host physiology, the gut microbiome, and environmental factors such as diet and nutrition. Multiomic analyses may help to identify potential nutrient-associated omic predictors of IBD, allowing for the design of targeted dietary approaches for disease prevention and management. Objectives: Our objective was to apply the bioinformatics tool, Consolidated Analysis of Network Topology and Regression Elements (CANTARE), to an integrated multiomics dataset to generate nutrient-associated predictive models for IBD. Methods: We previously used a published data set of microbiome relative abundance (mb), untargeted metabolomics (met), and microbial-derived enzymes (e) in stool samples from 153 adults (IBD = 111, healthy control = 42) to build a network of cross-omic relationships that differed by IBD status. We now revisit this network to identify diet-associated predictive models of IBD using linear regression via the CANTARE workflow. Results: The network included 20 literature-supported nutrient-associated predictors across 3 subnetworks. We created 1 predictive model from each subnetwork. These models (M1, M2, and M3) contained 3, 4, and 11 predictors, respectively. Model performance was high, with area under the receiver operating characteristic curve of 0.87, 0.90, and 0.95 and pseudo- Conclusions: CANTARE provides an unbiased and comprehensive strategy that can integrate multiple omics to identify potential nutrient-associated predictors of IBD. Our models support the generation of hypotheses for follow-up targeted investigation in future dietary interventions for the management of IBD.