Development of an Accelerated Workflow for Parameterizing Kinetic Models of Metabolism

Gopalakrishnan, S. - Presenter, The Pennsylvania State University
Dash, S. - Presenter, The Pennsylvania State University
Maranas, C. D., The Pennsylvania State University
Kinetic models can simulate an organism’s metabolism under genetic perturbations by capturing the changes in metabolite pools and the impact on reaction fluxes in the form of regulations. The challenge of estimating kinetic parameters can be resolved using the Ensemble modeling (EM) approach which samples parameters consistent with a training dataset. However, the EM approach implemented using GA is computationally expensive. Thus, we have described here a gradient-based kinetic parameterization workflow that exploits the strengths and overcomes the shortcomings of the EM approach. First, the kinetic parameters are anchored to the steady-state metabolic flux distribution in the wild-type strain by exploiting the rate laws governed by mass-action kinetics relating fluxes through elementary reactions to sampled enzyme levels and kinetic parameters. Following this, a genetic perturbation is introduced by adjusting the total abundance of specific enzymes relative to the wild-type strain so as to reflect the in vivo genetic perturbation in the form of up/down-regulations or gene deletions. The steady-state fluxes in the mutant conditions are evaluated by decomposing the underlying system of bilinear equations into two linear sub-problems and iterating between them until convergence is achieved. Based on the deviation of predicted fluxes from measured data, kinetic parameters are updated using a Newtonian step until optimality is achieved. This accelerated procedure has been applied to a C. thermocellum core model (106 reactions, 72 metabolites, 674 elementary kinetic parameters, and 4 mutant conditions) and an E. coli near genome-scale model (319 reactions, 276 metabolites, 2536 elementary kinetic parameters and 7 mutant conditions). Parameterization was completed in 12 hours for C. thermocellum and 48 hours for E. coli, compared to a computational time requirement of 1 week and 6 weeks for C. thermocellum and E. coli, respectively using the GA-based procedure.