Section: New Results
Inference of metabolic networks
Participants : David Sherman [correspondant] , Razanne Issa, Pascal Durrens.
We are particularly interested in incremental modeling of metabolic networks, where the target organism to be modeled is demonstrably similiar to other organisms for which whole or partial models are available. The other organisms are typically strains of the same species as the target, or species with a close phylogenetic relation to the target species. The similarity is measured genomically at different scales: sequence polymorphisms, expansions and contractions in conserved protein families, and genome rearrangements. We have defined and refined two complementary methods for inferring metabolic models for target species.
In the same way that comparative analysis of genomes and proteomes makes it possible to define protein families that summarize protein-coding genes into phyletic patterns [24] , comparative analysis of related metabolic models makes it possible to define network generalizations [26] that factor families of reactions and metabolites into summary graphs that preserve stoichiometry. These summaries can be used for expert curation and visualization [5] . An online demonstration tool is made available at http://mimoza.bordeaux.inria.fr/ .
Starting from an existing reference metabolic network and measures of similarity between the reference and the target organisms' genomes, we can use knowledge-based inference to rewrite the reference network based on these differences, and thus obtain a draft network for the metabolisms of the target organism [2] . This rewriting, formalized in the Pantograph system, can be extended to an abductive logic framework as described in Razanne Issa's thesis [19] . Current work aims at extending the Pantograph and ab-Pantograph frameworks to leverage reaction classifications obtained by network generalization.