Report on structural comparison and validation of the genome scale metabolic models for Saccharomyces cerevisiae

One of the objectives in the Unicellsys project is to develop computational tools to automatically modify the metabolic pathway models based on the experimental data and the bioinformatics information available. The machine learning tool could be developed within the framework of constraint-based optimization and/ or inductive logic programming (ILP). Meanwhile a good starting point for the yeast metabolic pathway model is important for the quality of the modified model.

There are a couple of existing genome scale reconstructed metabolic models for Saccharomyces cerevisiae from different research groups and of different revisions. Here we focus on the models from the latest reconstruction efforts, which include the Aber model [2, 3], the iIN800 model [7] and the consensus model [1].

The Aber model, which is a logical model for the yeast metabolism, has been constructed by RobotScientist's group at Aberystwyth University, UK. The iIN800 model came out of an international joint work and was constructed as a scaffold to query the lipid metabolism. The consensus model came from a latest curation effort by YSBN consortium in 'Jamboree' style, which was mainly conducted at the University of Manchester, UK.

In this work, we initially compared the structural information between the Aber model and the consensus model, and tried to unify the two if possible. Logical model simulation for single and double gene deletion were performed based on the two network models and results were validated by the existing experimental data. Furthermore, ux balance analyses have been performed utilizing the consensus model. Flux range analysis has been used to identify the network gaps including blocked reactions and dead-end metabolites.