CANTARE: finding and visualizing network-based multi-omic predictive models.

Area of Science:

• Computational biology and bioinformatics
• Systems biology
• Genomics and multi-omics research

Background:

• Multi-omic studies aim to identify predictive models using analytes from various biological 'omes'.
• Current analytical methods often analyze one 'ome' at a time or rely on pairwise correlations, limiting cohesive insights.
• There is a need for integrated approaches to build interpretable multi-omic predictive models.

Purpose of the Study:

• To develop a novel workflow for building predictive regression models from multi-omic network neighborhoods.
• To introduce CANTARE (Consolidated Analysis of Network Topology And Regression Elements) for analyzing multi-omic data.
• To enable the generation of interpretable, multi-omic predictive models.

Main Methods:

• CANTARE generates pairwise regression models across all analyte pairs from all 'omes', creating a network of relationships.
• Predictive logistic regression models are then built using analytes within network neighborhoods of interest.
• The workflow integrates network topology with regression analysis for consolidated multi-omic modeling.

Main Results:

• CANTARE was applied to multi-omic data (gut microbiome, metabolomics, microbial enzymes) from healthy controls and IBD patients.
• The method identified 8 unique predictive models with AUC > 0.90, using 3 to 13 predictors.
• CANTARE models demonstrated competitive AUC values, a greater dynamic range of predicted probabilities, and a higher likelihood of prioritizing multi-omic predictors compared to random forests and penalized regressions.

Conclusions:

• CANTARE provides a flexible framework for constructing parsimonious and interpretable multi-omic models.
• The models offer quantitative and directional effect sizes, aiding in hypothesis generation.
• This approach facilitates deeper biological insight and discovery from integrated 'omic' data.