AbstractWhile new genomes are sequenced at ever increasing rates, their phenotypic analysis remains a major bottleneck of biomedical research. The generation of genome-scale metabolic models capable of accurate phenotypic predictions is a labor-intensive endeavor; accordingly, such models are available for only a small percentage of sequenced species. The standard metabolic reconstruction process starts from a (semi-)automatically generated draft model, which is then refined through extensive manual curation. Here, we present a novel strategy suitable for full automation, which exploits high-throughput gene knockout or nutritional growth data. We test this strategy by reconstructing accurate genome-scale metabolic models for three strains ofStreptococcus, a major human pathogen. The resulting models contain a lower proportion of reactions unsupported by genomic evidence than the most widely usedE. colimodel, but reach the same accuracy in terms of knockout prediction. We confirm the models’ predictive power by analyzing experimental data for auxotrophy, additional nutritional environments, and double gene knockouts, and we generate a list of potential drug targets. Our results demonstrate the feasibility of reconstructing high-quality genome-scale metabolic models from high-throughput data, a strategy that promises to massively accelerate the exploration of metabolic phenotypes.Significance statementReading bacterial genomes has become a cheap, standard laboratory procedure. A genome by itself, however, is of little information value – we need a way to translate its abstract letter sequence into a model that describes the capabilities of its carrier. Until now, this endeavor required months of manual work by experts. Here, we show how this process can be automated by utilizing high-throughput experimental data. We use our novel strategy to generate highly accurate metabolic models for three strains ofStreptococcus, a major threat to human health.
Automated high-quality reconstruction of metabolic networks from high-throughput data