Enhancing genetic algorithm-based genome-scale metabolic network curation efficiency

Genome-scale metabolic modeling using constraint-based analysis is a powerful modeling paradigm for simulating metabolic networks. Models are generated via inference from genome annotations. However, errors in the annotation or the identity of a gene's function could lead to "metabolic inconsistency" rendering simulations infeasible. Uncovering the source of metabolic inconsistency is non-trivial due to network size and complexity. Recently published work uses genetic algorithms for curation by generating pools of models with randomly relaxed mass balance constraints. Models are evolved that allow feasible simulation while minimizing the number of constraints relaxed. Relaxed constraints represent metabolites likely to be the root of metabolic inconsistency. Although effective, the approach can result in numerous false positives. Here we present a strategy, MassChecker, which evaluates all of the relaxed mass balance constraints in each generation prior to the next round of evolution to determine if they had become consistent due to recombination/mutation. If so, these constraints are enforced. This approach was applied to the development of genome-scale metabolic model of B. anthracis. The model consisted of 1,049 reactions and 1,003 metabolites. The result was a 60% reduction in the number of relaxed mass balance constraints, significantly speeding up the curation process.