Analysis of the PROM algorithm as a tool to generate genome-scale metabolic-regulatory networks

Abstract

In this paper, we analyzed the capability of PROM’s algorithm to generate genome--‐scale metabolic--‐regulatory networks, which accurately predict growth phenotypes of transcriptional regulatory mutants under various conditions. E. coli, M. tuberculosis and S. cerevisase were used as model organisms. We showed that PROM could be successfully applied to model less complex systems (E. coli and M. tuberculosis) but not eukaryotes (S. cereavisae). The effects of the accuracy of the metabolic and regulatory networks reconstructions as well as the amount of gene expression data (microarrays) on PROM’s ability to simulate growth phenotypes was analyzed. It was determined that well defined metabolic model and transcriptional regulatory network were crucial for PROM to be predictive. However, accurately represented gene--‐ transcription factor (TF) interactions played a more significant role than the metabolic model. Also, those interactions had to be determined experimentally and not through an inference algorithm (such as ASTRIX). In case of the amount of gene expression data, it was observed that a number of microarrays needed for best PROM’s performance was species specific and incorporation of additional samples resulted in no further improvement of the model. The extension of PROM’s algorithm to predict changes in reaction rates (fluxes) for transcriptional regulatory mutants growing on different media showed that incorporation of Flux Variability Analysis (FVA) was not sufficient for such studies.