Probabilistic Metabolic Network Reconstruction: A Modular Approach to Integrate Omics Data and Represent Uncertainty

Metabolic network reconstruction is central to many aspects of metabolic analysis and engineering, and is essential for ongoing efforts towards mechanistic microbial ecosystem modeling. In general, the reconstruction process involves translating genomic information into a metabolic network that can provide insight into cell-wide metabolic fluxes and other phenotypes. A number of well-curated reconstructions exist for different model organisms, but a major challenge in the field is to reconstruct metabolic networks for diverse environmentally and medically relevant organisms. Various data sources including genomic, metabolomic, and phenotypic data can be used to guide metabolic network reconstruction. Additionally, there are several methods available for automatically generating draft reconstructions, including by using probabilistic genome annotations. However, few methods have directly attempted a comprehensive, automated integration of multiple data types, while at the same time explicitly dealing with the inherent uncertainty in the reconstruction process. We have developed a Markov chain Monte Carlo method to sample metabolic networks from a probability space shaped by the available data. Our method formalizes the data integration process and clarifies the representation of uncertainty by generating a natural ensemble of metabolic network reconstructions. As a proof of concept, we implemented this method on simulated genomic and metabolomic data from the E. coli iJO1366 metabolic network. Moving forward, we are working on including additional data types, such as phenotypic assays, with the goal of applying this method to a large group of environmentally-relevant marine microbes.