Analysing the Expressiveness of Metabolic Networks Representations

In this work, we explore the expressiveness of three graph-based representations of metabolic networks. We consider Abstract Metabolic Networks (AMNs), metabolic-Directed Acyclic Graphs (m-DAGs) and Reaction Graphs (RGs). These representations form a hierarchical view of the metabolism, AMNs being the most abstract, m-DAGs serving as the intermediate, and RGs being the most detailed. We evaluate their expressiveness for a case study comprising 331 Vertebrates and by using the Weisfeiler-Lehman graph kernel to perform the comparison. The results show that AMNs are not able to discern the various taxonomic groups at the Class level, while m-DAGs and RGs clearly distinguish Mammals, Fishes and Birds. When focusing on Mammals at the Order level, only m-DAGs are partially able to identify some of the taxonomic groups. Moreover, m-DAGs are able to distinguish Primates at the Infraorder level of taxonomy. Based on the obtained results, it emerges that m-DAGs are a good compromise between the amount of network information and the computational effort needed to obtain reliable patterns on the taxonomic clustering of the different organisms.