Bayesian Inference of Differential Metabolic Fluxes in Ulcerative Colitis using High-throughput Transcriptomics Data

Abstract

Advances in high-throughput technology have enabled the measurement of transcriptomic responses to a broad range of conditions. Our increased understanding of human metabolic reactions has also enabled the construction of large enzymatic network models. With the availability of detailed models and data to fit them to, various methods for analyzing the effects of transcriptional regulation on metabolism have been proposed. However, for conditions under which our understanding of the transcriptional response is limited, constructing a quantitative model for metabolic changes can prove challenging. In this work, we suggest a novel Bayesian approach that combines systems-level modeling of metabolism with Bayesian inference of the transcriptional response. Our method combines Bayesian reasoning with flux balance constraints that are used in mechanistic models of metabolism, and is thereby able to incorporate both uncertainty and systems-level metabolic knowledge into the model. It models only metabolic differences between two conditions, thereby simplifying the model to capture only effects that are derived from the transcriptomic response to a specific stimulus. We demonstrate the usefulness of our approach by studying the transcriptional regulation of metabolism in two independent ulcerative colitis datasets.