Bayesian inference of the sites of perturbations in metabolic pathways via Markov chain Monte Carlo

Motivation: Genetic modifications or pharmaceutical interventions can influence multiple sites in metabolic pathways, and often these are distant from the primary effect. In this regard, the ability to identify target and off-target effects of a specific compound or gene therapy is both a major challenge and critical in drug discovery. Results: We applied Markov Chain Monte Carlo (MCMC) for parameter estimation and perturbation identification in the kinetic modeling of metabolic pathways. Variability in the steady-state measurements in cells taken from a population can be caused by differences in initial conditions within the population, by variation of parameters among individuals and by possible measurement noise. MCMC-based parameter estimation is proposed as a method to help in inferring parameter distributions, taking into account uncertainties in the initial conditions and in the measurement data. The inferred parameter distributions are then used to predict changes in the network via a simple classification method. The proposed technique is applied to analyze changes in the pathways of pyruvate metabolism of mutants of Lactococcus lactis, based on previously published experimental data. Availability: MATLAB code used in the simulations is available from ftp://anonymous@dbkweb.mib.man.ac.uk/pub/Bioinformatics_BJ.zip Contact: bayujw@ieee.org Supplementary information: Supplementary data are available at Bioinformatics online.