Modelborgifier, a Tool for Semi-Automated Assimilation of Metabolic Reconstructions

Historically, genome-scale metabolic reconstructions come in a great variety of file formats and use different naming schemes for reactions and metabolites. While SBML is on good track to becomeing the standard file format, no generally accepted naming scheme has emerged, yet. However, a unified naming scheme greatly simplifies sharing of reaction- and metabolite-associated information across reconstructions, comparing and combining reconstructions, and integrating reconstructions with measured data.

We developed modelBorgifier^[1]to facilitate the process of assimilating reconstructions. In an automated first step, a pairwise comparison of all reactions and metabolites in 40 parameters such as ID, name, and stoichiometry is computed. This scoring is the basis for a GUI-guided user oversight of the semi-automated reaction- and metabolite matching. Machine learning is iteratively employed to enhance the automated scoring based on user input. ModelBorgifier is implemented in Matlab and integrates seamlessly with the popular openCOBRA toolbox. We demonstrate the advantage of assimilated models with examples of industrial applications of stoichiometric modeling. Moreover, we will present our latest achievements in improving the quality of stoichiomertric models.

[1] Sauls, J.T. and Buescher, J.M. (2014) Assimilating genome-scale metabolic reconstructions with modelBorgifier. Bioinformatics, 1–3. 
DOI: 10.1093/bioinformatics/btt747