Gap-filling analysis of the iJO1366 Escherichia coli metabolic network reconstruction for discovery of metabolic functions

A new model improvement workflow was developed to compare model based phenotypic predictions to experimental data to fill gaps and correct errors. A Keio Collection based dataset of E. coli gene essentiality was obtained from literature data and compared to model predictions. The SMILEY algorithm was then used to predict the most likely missing reactions in the reconstructed network, adding reactions from a KEGG based universal set of metabolic reactions. The feasibility of these putative reactions was determined by comparing updated versions of the model to the experimental dataset, and genes were predicted for the most feasible reactions.Numerous improvements to the iJO1366 metabolic reconstruction were suggested by these analyses. Experiments were performed to verify several computational predictions, including a new mechanism for growth on myo-inositol. The other predictions made in this study should be experimentally verifiable by similar means. Validating all of the predictions made here represents a substantial but important undertaking.Constraint-based modeling is a widely used systems biology method and is particularly well suited for predicting the phenotypes of microbial organisms after gene knockouts or when grown on different substrates [1-3]. These variable conditions are simply represented as additional constraints on a model, and growth can be predicted by flux balance analysis (FBA) [4]. Because not every realistic constraint is represented in a typical metabolic model, it is quite possible for such a model to predict growth under conditions where growth does not really occur. The actual organism may not express a required gene for growth, or fluxes may be limited by kinetic or thermodynamic constraints, for example. This case is called a false positive prediction. On the other hand, false predictions of no growth can be taken as indications that the model is missing an essential reaction [5]. This prediciton is called a false negative. No current m