Finding and visualizing network-based multi-omic predictive models

Background: One goal of multi-omic studies is to identify interpretable predictive models for outcomes of interest, with analytes drawn from multiple omes. Such findings support refined biological insight and hypothesis generation. However, standard analytical approaches are not designed to be "ome aware." Some researchers analyze data one ome at a time. Others resort to correlation studies, cataloging pairwise relationships, but lacking an approach for cohesive summaries of these catalogs. Methods: We present a novel workflow for building predictive models from multi-omic networks. First, we perform pairwise regressions across all pairs of analytes from all omes, en- coding the resulting "top table" in a network. To explore these results, we developed an interactive web application, VOLARE (Visualization of LineAr Regression Elements). Next, we generate multi-variable regression models using network neighborhoods. We call this method CANTARE (Consolidated Analysis of Network Topology And Regression Elements). The JavaScript web application, R scripts, and User ̓s Guides are available at https://sourceforge.net/projects/cytomelodics/. Results: We applied our workflow to two studies. In a primary analysis of data from a weight loss intervention (gut microbiome, DNA methylation, targeted metabolomics, and clinical outcomes), we identified predictive models for eight clinical outcomes with adjusted R² ranging from 0.34 to 0.78. Our models identified specific DNAme sites, microbes, and metabolites relevant to obesity. In a reanalysis of data from healthy controls and patients with inflammatory bowel disease (IBD) (gut microbiome, metabolomics, and microbial-derived enzymes), we identified 8 predictive models for IBD group having AUC> 0.90. We compared the CANTARE models to those generated by random forests and penalized regressions, analyzing AUC, predictions, and predictors. CANTARE AUC values were competitive with those generated by random forests and penalized regression. CANTARE models were significantly more likely to include predictors frommultiple omes than were the alternatives. Conclusion: VOLARE and CANTARE offer a flexible approach for building parsimonious, interpretable multi-omic models. These models yield quantitative and directional effect sizes and can inform follow-up investigations. In the case of the weight loss intervention, our analysis supports the feasibility of using baseline multi-omic features to provide insight for precision nutrition-based weight loss interventions.