Section: New Results

Improving inference of metabolic models

Participants : David James Sherman [correspondant] , Pascal Durrens, Razanne Issa, Anna Zhukova.

The Pantograph approach uses reference model annotated by gene associations, and voting between complementary predictions of homology between reference genes and target genes, to decide whether a reaction that is present in the scaffold ought be be present in the target. A gene association implicitly represents expert knowledge about the role of genes in a compact way. If the gene association can be rewritten into a possibly satisfiable formula, then the corresponding reaction is instantiated in the target model.

Historically, gene associations have been used intuitively by experts during the model design and curation process, and are often inconsistent. We have formalized the construction of gene associations based on the semantics of different interpretations, showing how different boolean formulas should be constructed when the application is i) metabolic model inference, ii) flux-balance analysis, iii) hierarchical modeling, or iv) dynamic simulation (Razanne Issa, MS in prep.).

Second, we have refined our strategy for inferring metabolic models using abductive logic. We have shown that given a set of genes as observations in the target organism, and rules for rewriting gene associations while respecting integrity constraints for the model, then the reactions in the target model can be abduced as hypotheses that “explain” the presence of a maximial number of genes in the target genome. The advantage of this approach is that it can invent, through specialization, reactions that are not present per se in the reference model. Two classes of reactions can be invented: substrate-specific reactions inferred from expansion in gene families, and transport reactions needed to maintain model integrity for constitutive compartments.