Distilling Mechanistic Models From Multi-Omics Data

Abstract

High-dimensional multi-omics data sets are increasingly accessible and now routinely being generated as part of medical and biological experiments. However, the ability to infer mechanisms of these data remains low due to the abundance of confounding data. The gap between data generation and interpretation highlights the need for strategies to harmonize and distill complex multi-omics data sets into concise, mechanistic descriptions. To this end, a four step analysis approach for multiomics data is herein demonstrated, comprising: filling missing data and harmonizing data sources, inducing sparsity, developing mechanistic models, and interpretation. This strategy is employed to generate a parsimonious mechanistic model from high-dimensional transcriptomics and metabolomics data collected from a murine model of Clostridioides difficile infection. This approach highlighted the role of the Stickland reactor in the production of toxins during infection, in agreement with recent literature. The methodology present here is demonstrated to be feasible for interpreting multi-omics data sets and it, to the authors knowledge, one of the first reports of a successful implementation of such a strategy.