Classifying metabolic networks using physically-motived distance metrics

Comparative genomics has been essential to inferring the evolutionary trajectories of species and producing classifications on the basis of DNA. However, purely genomic classifications present an incomplete characterization of the functional role of species in an ecological environment. Previous work has shown the potential for classification methods using metabolic networks but these do not utilize the full topological information contained in the networks. Here we introduce a physically-motivated computational framework for classifying organisms which compares matrix representations of metabolic networks to compute distances between fundamental subspaces of stoichiometric matrices: the spaces of steady-state reaction fluxes and time-invariant metabolite pools. We find that for the AGORA2 human gut microbiome dataset these metrics produce distances that are inequivalent to those produced by a phylogenetic analysis reflecting an inability of the genetic information to completely produce metabolic classifications. We also demonstrate that in silico genetic knockouts of common bacterial laboratory strains are generally proximal to each other, highlighting the robustness of the flux and pool spaces to genetic perturbation.