(612f) A Network Reduction Tool for Compressing and Visualizing Genome-Scale Metabolic Models

Flux balance analysis (FBA) is an in-silico method for predicting intracellular reaction rates (fluxes) throughout a metabolic network. In FBA, a metabolic network modeled by a stoichiometric matrix is subjected to linear optimization with a suitable objective function and constraints. Results of such analyses are difficult to readily interpret because genome-scale metabolic models used in FBA can contain thousands of reactions. For instance, models containing ~5,000 reactions have been recently generated (Gilbert et al. Briefings in Bioinformatics, doi: 10.1093/bib/bbx096, 2017; unpublished results from our research group). The size of the stoichiometric matrix and the flux vector resulting from such large models makes it nearly impossible to visualize or meaningfully interpret the results of their FBA simulations. Manual attempts to parse the results to find the fluxes and metabolites important to the user have limited utility. This problem is further compounded in FBA problems where two or more metabolic scenarios are compared. Thus, a computational method that automatically reduces the FBA results to a smaller network is highly desirable. Such a method should produce a subnetwork that is representative of the original genome-scale network, but also small enough so that it affords easy visual interpretation. Towards this goal, researchers have previously developed optimization-based methods to reduce FBA models; however, these methods act on the FBA model before the simulation results are generated. Thus, they are of a black-box nature and information about the network could often be lost. Contrastingly, we have developed a network reduction methodology that is optimization-free and based solely on linear algebraic operations and tools of convex analysis. This method acts on the results of FBA, post-simulation, and compacts the network. This allows the highlighting of fluxes through metabolic nodes deemed of interest to the user. This enables end users to more easily garner the important motifs of the network without sacrificing the crucial information about the metabolic behavior of an organism that can only be obtained by simulating a genome-scale model. Additionally, the reduced network can compare the overall conversions of chosen metabolic pathways , which has important implications for validating FBA simulations with experiments. In this presentation, we will delineate the network reduction process, explain the novelty of our methodology and provide some illustrative results on genome-scale metabolic models. These results will include those developed in our research group, particularly a genome-scale model of poplar metabolism and ATP synthesis in maize embryos. We will also report the use of our network reduction methodology to compare FBA simulations with previously reported experimental results (Zhang et al. The Plant Journal 93, 472-488, 2018).