High-throughput generation and optimization of genome-scale metabolic models

2 Genome-scale metabolic models have proven to be crucial resources for translating detailed knowledge of thousands of distinct biochemical processes into global predictions of organism behavior. These models can be used to predict essential genes, organism phenotypes, organism response to mutations, and metabolic engineering strategies [1]. The models also serve as platforms for assessing and expanding knowledge of metabolism via an iterative cycle of experimentation, prediction, and reconciliation [2]. Despite these many applications, methods for creating genome-scale models are failing to keep pace with genome sequencing. In the past decade, 800+ prokaryotic genomes have been submitted to NCBI, but only 30 genome-scale models have been published [3]. To address this problem, we have developed the High-Throughput Genome-scale Metabolic Reconstruction (HT-GMR) pipeline, which rapidly generates predictive genome-scale metabolic models from prokaryotic genome sequences. This pipeline integrates numerous technologies for automating portions of the reconstruction process with minimal annotation and assembly [6], thermodynamic analysis to determine reaction reversibility [7, 8], and model optimization to fit experimental data [8-10]. We used the HT-GMR pipeline to generate 130 new genome-scale metabolic models and fit 22 of these models to available experimental data [11-18]. Gibbs free energy of reaction values were generated for 90% of the reactions in every model [7]. Any gaps preventing models from growing on known minimal media were identified and filled to enable the prediction of phenotypes and essential gene sets. of 19 cases, the HT-GMR models include more genes than their published counterparts. Validation of the 22 models with available growth phenotype data [11-18] reveals the models to have an average accuracy of 66% before optimization 3 and 87% after optimization, which closely approaches the accuracy of available published models. Until now, the genome-scale metabolic reconstruction process has followed a " one genome at a time " paradigm, where years of manual effort have been expended to build the most comprehensive model possible for a single organism. However, this paradigm does not scale in a world where sequencing capacity is rising and sequencing cost is falling at exponential rates; high-throughput methods are needed for performing genome-scale metabolic reconstruction without sacrificing quality or accuracy. To meet this challenge, we have designed and implemented the High-Throughput Genome-scale Metabolic Reconstruction (HT-GMR) pipeline within the SEED framework for genome annotation and analysis [5]. The SEED framework addresses two fundamental needs for rapid generation of accurate genome-scale metabolic models: the need for high-quality, consistent underlying …