Discovering subnetworks in SBML models.

Abstract

Motivation: Many advances in biomedical research are driven by structural analysis, which investigates interconnections between elements in biological systems (e.g. structural analysis of proteins to infer their function). Herein, we consider subnet discovery in chemical reaction networks (CRNs)-discovering a subset of a target CRN, i.e. structurally identical to a reference CRN. Structural analysis techniques such as motif finding and graph mining look for small, arbitrary, and commonly occurring substructures (e.g. three gene feedforward loops). In contrast, subnet discovery looks for larger, specific, and infrequently occurring substructures (e.g. 10 reactions mitogen-activated protein kinase (MAPK) pathway).

Results: We introduce pySubnetSB, an open source Python package for discovering subnets in CRNs that are represented in the Systems Biology Markup Language (SBML) community standard. We show that pySubnetSB achieves large reductions in computational complexity for subnet discovery. For example, in studies of randomly selected target networks with 100 reactions each with a random reference network with 20 reactions, computations are reduced from an infeasible 1078 evaluations to a more practical 108 evaluations. We develop a methodology for assessing the statistical significance of subnet discovery. Last, we study subnets in BioModels for approximately 200 000 pairs of reference and target models. We show that for a reference MAPK pathway, subnet discovery correctly indicates the presence of MAPK function in several target models. The studies also suggest two interesting hypotheses: (a) the potential presence of hidden oscillators in several models in BioModels, and (b) the possibility of a conserved mechanism for intracellular immune response.

Availability and implenetation: pySubnetSB is installed using pip install pySubnetSB, and is hosted at https://github.com/ModelEngineering/pySubnetSB/.