A Novel Approach to Characterization and Reduction of Uncertainty in the Kinetic Models of Large-Scale and Genome-Scale Metabolic Networks

  A Novel Approach to Characterization and Reduction of Uncertainty

  In the Kinetic Models of Large-scale and Genome-scale Metabolic Networks

Stefano Andreozzi^{1, 2}, Ljubisa Miskovic^{1, 2} and Vassily Hatzimanikatis^1,2

¹Laboratory of Computational Systems Biotechnology, Ecole Polytechnique Federale de Lausanne (EPFL), CH 1015 Lausanne, Switzerland,

²Swiss Institute of Bioinformatics (SIB)

Abstract

Accurate determination of physiological states of cellular metabolism requires detailed information about metabolic fluxes, metabolite concentrations and distribution of enzyme states. Integration of fluxomics and metabolomics data, and thermodynamics-based metabolic flux analysis contribute to improved understanding of steady-state properties of metabolism. Although knowledge about kinetics and enzyme saturations is essential for quantitative understanding of metabolic dynamics, it remains scarce and involves uncertainty. In this work, we present a novel computational methodology that makes use of machine learning, systems engineering concepts and Monte Carlo sampling techniques allowing us to determine and quantify the enzyme saturations that correspond to a given metabolic flux and a given metabolite concentration vector. We applied the proposed method to compute the distribution of enzyme saturations for a physiology of glucose-fed E. coli producing 1,4-Butanediol. Surprisingly, we discovered that the observed physiological state corresponds to a restrained range of saturations of only a few enzymes. Furthermore, this analysis indicated which are the enzymes that should be manipulated in order to engineer the reference state of the cell in a desired way. The proposed approach offers a new way to guide experiments for the identification of key metabolite concentrations and enzyme saturations by identifying in advance which quantities should be measured to better characterize a particular physiological state. Moreover, the proposed approach sets up the foundations of a novel type of approaches for efficient, non-asymptotic, uniform sampling of solution spaces.