(317d) Inference of Putative Metabolite-Dependent Regulatory Interactions from Metabolomics Data

Despite providing one of the most direct readouts of cell phenotype available, to date there has been surprisingly little application of metabolomics (the systems-level measurement and analysis of metabolites, which are the small molecules that supply and drive cellular processes) to metabolic engineering strain design. Similarly, very few of the computational methods that have been commonly used for strain design, such as constraint-based modeling strategies, take advantage of metabolomics. In order to address this gap, we are developing new computational strategies that use dynamic metabolomics data to improve the accuracy of strains designed in silico.

Metabolite-dependent regulatory interactions, such as allosteric regulation, are not as well- studied as other types of regulation (e.g., transcriptional regulation). Here, we use metabolomics data and existing knowledge of metabolic network structures to infer putative regulatory interactions via structure learning algorithms including Bayesian Networks. To infer these interactions more directly, we use dynamic flux estimation and metabolomics data to generate a flux distribution complementary to the metabolite abundances; the structure learning is performed on this augmented dataset. The resulting network interactions can be sorted by likelihood and used as a list of putative interactions which can then be tested by adding them to a metabolic model and assessing whether their presence is beneficial or detrimental to model accuracy.