Enhanced Metabolic Discovery Using Enzyme Family Inference and Approximate Bayesian Computation

Enhanced metabolic discovery using enzyme family inference and Approximate Bayesian Computation

Authors: William A Bryant, John W Pinney and Michael JE Sternberg

Many bacterial genome-scale metabolic models (GEMs) have been published in the last decade and increasing numbers are being produced month-by-month.  Every one of these GEMs contains information gleaned from intensive manual curation, and the gene-protein-reaction (GPR) relationships in those models represent an excellent source of enzymatic knowledge that could be used to infer new enzymatic functions in other bacteria.  A small subset of these data has been used to infer new GPR relationships in a limited number of cases [1], by using BLAST comparisons to infer gene orthology.  We present a new fully computational technique for postulating novel GPRs in a model under development based on precalculated protein family membership (COGs [2]), which enables GPR inference across a huge range of bacteria, finding candidates for both addition to and removal from the model based on a large number of curated metabolic models.  Further we present an Approximate Bayesian Computational approach based on a simulated Monte Carlo method (ABC-SMC [3]) which determines the most likely subset of additions and removals to improve such a model, based on the congruence between experimental and computational predictions of bacterial growth (usually gene essentiality information).  The ABC approach makes use of a prior probability for the inclusion/exclusion of each specific inferred reaction/enzyme link (REL) and while this can be flat (the same for every REL), it can also be used to take into account such information as operon membership of the constituent genes of the enzyme, gene co-expression and any other information determining gene functional relatedness.  Ongoing validation shows excellent recapitulation of many of the enzymes in a curated model of M. tuberculosis (iNJ661 [4]) and it shows great promise for accelerating the curation of metabolic models as well as the ability to posit and validate new functionality and identify errors in published metabolic models.

References:

[1]      R. A. Notebaart, F. H. J. van Enckevort, C. Francke, R. J. Siezen, and B. Teusink, “Accelerating the reconstruction of genome-scale metabolic networks.,” BMC Bioinformatics, vol. 7, no. 1, p. 296, Jun. 2006.

[2]      J. Muller, D. Szklarczyk, P. Julien, I. Letunic, A. Roth, M. Kuhn, S. Powell, C. von Mering, T. Doerks, L. J. Jensen, and P. Bork, “eggNOG v2.0: extending the evolutionary genealogy of genes with enhanced non-supervised orthologous groups, species and functional annotations.,” Nucleic Acids Research, vol. 38, no. Database issue, pp. D190–5, Jan. 2010.

[3]      T. Toni, D. Welch, N. Strelkowa, A. Ipsen, and M. P. H. Stumpf, “Approximate Bayesian computation scheme for parameter inference and model selection in dynamical systems.,” J R Soc Interface, vol. 6, no. 31, pp. 187–202, Feb. 2009.

[4]      N. Jamshidi and B. Ø. Palsson, “Investigating the metabolic capabilities of Mycobacterium tuberculosis H37Rv using the in silico strain iNJ661 and proposing alternative drug targets,” BMC Syst Biol, vol. 1, no. 1, p. 26, 2007.