Path2Models: large-scale generation of computational models from biochemical pathway maps

Abstract

Background: Systems biology projects and omics technologies have led to a growing number of biochemical pathway models and reconstructions. However, the majority of these models are still created de novo, based on literature mining and the manual processing of pathway data.

Results: To increase the efficiency of model creation, the Path2Models project has automatically generated mathematical models from pathway representations using a suite of freely available software. Data sources include KEGG, BioCarta, MetaCyc and SABIO-RK. Depending on the source data, three types of models are provided: kinetic, logical and constraint-based. Models from over 2 600 organisms are encoded consistently in SBML, and are made freely available through BioModels Database at http://www.ebi.ac.uk/biomodels-main/path2models. Each model contains the list of participants, their interactions, the relevant mathematical constructs, and initial parameter values. Most models are also available as easy-to-understand graphical SBGN maps.

Conclusions: To date, the project has resulted in more than 140 000 freely available models. Such a resource can tremendously accelerate the development of mathematical models by providing initial starting models for simulation and analysis, which can be subsequently curated and further parameterized.

Figure 1

Workflow leading from pathway descriptions to computational models. From the pathway databases on the left, information is extracted and encoded in SBML. Mathematical features, such as kinetic rate equations and flux bounds, are then added to each model, along with a graphical description. The completed models are all distributed through the BioModels Database. See Methods for a detailed explanation of each step.

Figure 2

SBGN Process Description map of a pathway, cutout of the pathway and parts of the SBML file describing the reactions shown in the cutout.

Figure 3

Rate equations from SABIO-RK for models from selected organisms.

Figure 4

Workflow indicating the SuBliMinaL Toolbox modules that were linked to produce draft metabolic models from the source data.KEGG extract and MetaCyc extract produce MIRIAM-annotated SBML representations of the contents of KEGG and MetaCyc, respectively. Metabolite and reaction ids are reconciled through reference to the MNXref namespace, unifying the metabolites to an assumed intracellular pH of 7.3, and mass and charge balancing reactions where possible. The Merge module merges the individual reconstructions from KEGG and MetaCyc, to which a limited growth medium and transport reactions are added, along with gene-protein relationships (GPRs) and flux bounds. The models are then formatted to allow for their analysis with the COBRA Toolbox and then released as draft models that represent the union of the information held in both KEGG and MetaCyc.

Figure 5

Phylogenetic tree illustrating all 2 630 genome-scale metabolic models. The tree is color coded, indicating the presence of archaea, bacteria and eukaryota in the collection. Analysis results of each model are displayed, with bars indicating the number of metabolic reactions, metabolites, makeable metabolites and makeable biomass components in blue, red, purple and green respectively. In this illustration, the bars have been scaled for ease of visualization.

Figure 6

A zoomed in view of the eukaryotic branch of the phylogenetic tree of Figure5. The online iTOL web application version of the tree, available at [40], allows for zooming, searching and visualization of the tree and its associated statistics.

Figure 7

Distribution of the models generated by the project according to their size, in terms of the number of molecular species (blue) and the number of mathematical relationships – i.e. reactions, transitions, rules etc. (salmon) in each class. A-C: the whole genome reconstructions, qualitative models, and chemical kinetic models. D-E: the curated and non-curated literature-based branches of the BioModels Database.

Figure 8

Relative sizes of the different classes of models, based on their main Gene Ontology (GO) annotations. The GO terms annotating the SBML Model element for each model generated by the project were collected, and clustered to generate groups of models covering (what are considered therefrom to be) the same domain of biology.