PaperlayerView exact revision on arXiv

Source-linked AI summary

Large-scale generation of computational models from biochemical pathway maps

Finja Büchel, Nicolas Rodriguez, Neil Swainston, Clemens Wrzodek, Tobias Czauderna, Roland Keller, Florian Mittag, Michael Schubert, Mihai Glont, Martin Golebiewski, Martijn van Iersel, Sarah Keating, Matthias Rall, Michael Wybrow, Henning Hermjakob, Michael Hucka, Douglas B. Kell, Wolfgang Müller, Pedro Mendes, Andreas Zell, Claudine Chaouiya, Julio Saez-Rodriguez, Falk Schreiber, Camille Laibe, Andreas Dräger, Nicolas Le Novère

arXiv:1307.7005v2q-bio.MN

TL;DR

Mathematical model construction from pathway data remains labor-intensive, motivating automated reconstruction workflows. This work systematically generates standardized models from biochemical pathways, producing a large collection that serves as a starting point for curation and comparison.

  • Problem

    Genome-scale metabolic reconstructions commonly require labor-intensive manual development, with reported production timescales of up to two years.

  • Method

    The workflow combines pathway information with experimentally determined kinetic laws and parameters where available, while supporting freely available, standardized model-building software.

  • Results

    The project produced the largest collection of genome-scale metabolic reconstructions to date, with models available for subsequent curation and systematic comparison across species.

  • Takeaways & Limitations

    Common formatting, identifiers, and semantic annotations make the generated models useful starting points for manual or semi-automated curation and for comparing metabolism across species.

  • Takeaways & Limitations

    KEGG signaling models may contain only topological relationships and interaction signs, so they remain scaffolds requiring parameterization before simulation.

Abstract

from arXiv · show

Background: Systems biology projects and omics technologies have led to a growing number of biochemical pathway reconstructions. However, mathematical models are still most often created de novo, based on reading the literature and processing pathway data manually. Results: To increase the efficiency with which such models can be created, we automatically generated mathematical models from pathway representations using a suite of freely available software. We produced models that combine data from KEGG PATHWAY, BioCarta, MetaCyc and SABIO-RK; According to the source data, three types of models are provided: kinetic, logical and constraint-based. All models are encoded using SBML Core and Qual packages, and available through BioModels Database. Each model contains the list of participants, the interactions, and the relevant mathematical constructs, but, in most cases, no meaningful parameter values. Most models are also available as easy to understand graphical SBGN maps. Conclusions: to date, the project has resulted in more than 140000 models freely available. We believe this resource can tremendously accelerate the development of mathematical models by providing initial starting points ready for parametrization.

BACKGROUND

Pathway databases capture biochemical interactions but are largely qualitative and static, while quantitative models require substantial manual construction and parameter information. Path2Models addresses this bottleneck by generating consistently formatted computational models at scale for community reuse.

  • Pathways represent chains of molecular interactions and reactions that alter the concentration, state, or location of chemical entities.
  • Public pathway resources are useful for mapping experimental data but generally provide qualitative, static representations rather than quantitative cellular models.
  • Quantitative modeling requires reaction knowledge, stoichiometric and kinetic parameters, and often initial metabolite and enzyme concentrations, which are frequently unavailable.
  • Genome-scale metabolic reconstructions can require labor-intensive manual development lasting up to two years.
  • Path2Models generates models from pathways on a large scale using consistent formats and makes them available to the community.
  • The project produces kinetic, logical, and genome-scale models encoded in SBML, often paired with SBGN representations and semantic annotations for reuse and integration.

RESULTS

The project uses a multi-stage pipeline to convert pathway data into standardized model types. It combines process-based kinetic modeling, qualitative influence modeling, and genome-scale reconstruction, with selected kinetic information added from SABIO-RK.

  • A software pipeline converts pathway data into a standard computer-readable format before subsequent processing steps.
  • Three parallel pipelines generate kinetic metabolic models, qualitative influence models, and genome-scale metabolic reconstructions.
  • 112 898 KEGG maps covering up to 154 metabolic pathways in 1 514 organisms were converted into SBML Level 3 Core process-description models.
  • 6 204 reactions, representing 0.22% across the considered organisms, received rate laws from SABIO-RK; missing rate laws were inferred ab initio.

Generation of qualitative models from signaling pathways

The project converts non-metabolic pathway data into SBML Qual influence models and integrates pathway information across resources for standardized reconstruction. These models support further analysis and curation but may remain incomplete or underdetermined.

  • Generation of qualitative models from signaling pathways: 27 306 KEGG maps describing 167 non-metabolic pathways in 1 514 organisms were converted into SBML Level 3 Qual influence models.
  • Generation of qualitative models from signaling pathways: Applying SBML Qual to actual pathways exposed specification problems and contributed to finalizing the package.
  • Generation of qualitative models from signaling pathways: A sign attribute was introduced to mark positive, negative, or unknown interaction effects and constrain later logical parameterization.
  • Generation of qualitative models from signaling pathways: Some KEGG modification relations cannot determine positive or negative influence, so their interaction sign is set to unknown when necessary.
  • Genome-scale metabolic reconstructions: 2 630 genome-scale reconstructions combined KEGG and, where available, MetaCyc data while reconciling identifiers through MNXref.
  • Genome-scale metabolic reconstructions: Only the Drosophila melanogaster model synthesized all 30 specified biomass components; the human model showed several amino-acid synthesis anomalies requiring curation.
  • Genome-scale metabolic reconstructions: The complete model collection is listed in Supplementary Table 1 and visualized through a phylogenetic tree.

Access to the resulting knowledge base

The generated models are publicly distributed through BioModels Database and provide reusable starting points for biological modeling. Their usefulness depends on the completeness and accuracy of the pathway sources, while model sizes and connectivity vary across model types.

  • Access through BioModels Database: The Path2Models branch of BioModels Database distributes the generated models in standard formats with free licensing and dedicated search infrastructure.The resource is intended for retrieval, reuse, and continued revision rather than as a frozen collection.
  • Use as starting points: The workflow shifts bulk pathway processing away from repeated manual effort, allowing scientists to add parameters and initial conditions and run simulations.The automatically generated models are starting points for question-specific adaptation and analysis.
  • Quality boundary: Model quality depends on the accuracy and completeness of source pathway information, which automation cannot fully repair when biological information is missing.Missing compartment localization can yield a single compartment containing all molecular species.
  • Model structure: Individual metabolic pathway models are often underdetermined, with more state variables than mathematical relationships, unlike whole-genome reconstructions and curated simulation models.The text attributes this pattern possibly to missing reactions caused by homology-based inference in KEGG pathways.
  • Scale and reuse: The project extends earlier automation efforts by systematically applying model reconstruction across a wide range of organisms and producing a large comparative collection.Common formatting, identifiers, and semantic annotations support later curation and cross-species comparison.

Complementing pathway models with kinetic information

The workflow supplements sparse experimental kinetics with generic rate laws and produces partially parameterized models. Logical signaling models may remain topological scaffolds when pathway sources lack mechanistic detail.

  • Kinetic information: Kinetic data cover only 12.2% of known metabolic reactions in Homo sapiens, motivating generic rate laws for incompletely characterized pathways.The passage notes that coverage is lower for other organisms.
  • Kinetic information: The workflow selects the simplest modular rate-law version, cat, for reversible reactions as a parsimonious modeling choice.The available versions increase in complexity from explicit cat to Haldane-compliant hal and Wegscheider-compliant weg forms.
  • Kinetic information: Because modular rate laws apply only to reversible metabolic reactions, additional generic rate equations are required for the large-scale workflow.Generic rate laws remain necessary even if experimentally determined rate laws become more available.
  • Parameterization: The automatically generated models are only partially parameterized, so they do not generally provide complete simulation-ready quantitative models.Parameter and initial-condition completion remains outside the automated output described here.
  • Logical models: KEGG signaling models without mechanistic details contain topological relationships and interaction signs but no logical rules for combined effects.They are scaffolds requiring default functions, manual refinement, or experimental data before simulation.

Creation of SBGN maps applying constraint-based layout

The project provides models with SBGN visual representations and uses source pathway layouts to improve map readability. Its freely available software supports extension to additional pathway resources and reuse of generated models.

  • SBGN visualization: SBGN maps provide a uniform graphical representation intended to facilitate human understanding of the generated models.The workflow uses initial element positions extracted from KEGG graphical pathway representations to improve layout readability.
  • Workflow reuse: All software building blocks are freely available, enabling similar workflows to incorporate pathway information from additional databases.The resulting models are added to BioModels Database for retrieval and reuse.
  • Workflow reuse: The workflow can produce models at scales where manual reconstruction is hardly an option.This supports automated development as pathway generation from genomic and metagenomic data becomes common.
  • Source data: The workflow obtains KEGG pathway data in KGML format, including pathway entries and interactions used for model construction.The downloaded collection included generic, reference, and organism-specific pathways for 1 515 species.
  • Source data: Genome-scale reconstructions use KEGG and MetaCyc data, with MNXref reconciling metabolite and reaction identifiers across resources.This construction used a more recent KEGG web-service version than the quantitative and qualitative model workflow.

Generation of SBML Level 3 Core from KEGG metabolic pathways

KEGG pathway entries were translated into SBML Level 3 Core models with annotated species, reactions, and automatically assigned kinetic laws. Missing laws were generated using selected rate-law families and initialized with consistent default parameters and units.

  • Each KGML entry became an SBML Level 3 species with an assigned SBO term, while each reaction preserved participants, reversibility, and stoichiometry.
  • SABIO-RK rate laws and kinetic parameters were extracted when matching reactions were identified through MIRIAM-compliant annotations.
  • Missing rate laws were generated with SBMLsqueezer using the Common Modular rate law for reversible enzyme-catalyzed reactions.The explicit cat form was selected because it requires fewer independent parameters than the Haldane- and Wegscheider-compliant forms.
  • Modular rate-law structure varies with reactants, products, and modifiers, while the cat, Haldane-compliant, and Wegscheider-compliant versions determine the numerator.The cat version has the smallest number of parameters.
  • Irreversible reactions used default laws including Henri-Michaelis-Menten, random-order ternary-complex, and convenience rate laws, depending on the reaction case.
  • Gene-regulatory processes used generalized Hill equations, genes were treated as constant pools, and newly generated parameters and initial amounts were set to 1.0.Default substance, time, and volume units were mole, second, and litre when prior units were unavailable.

Development and implementation of SBML Level 3 Qual package

SBML Level 3 Qual extends shared SBML Core with constructs for qualitative models, including logical and qualitative Petri-net representations. It encodes discrete activities and their transformations with QualitativeSpecies and Transition elements.

  • SBML Level 3 uses a shared Core package plus domain-specific packages that add representational features.
  • The qual package represents qualitative models such as logical and qualitative Petri-net models.
  • QualitativeSpecies encode entity-associated variables that can take discrete Boolean or multi-valued activity levels.
  • Transition elements encode rules governing how output nodes evolve depending on input-node states.

Generation of SBML Level 3 Qual from KEGG signaling pathways

KEGG signaling and other non-reaction relations were converted into SBML qual models, with BioCarta enrichment assigning activating or inhibiting subtypes. The broader pipeline also produced genome-scale reconstructions using KEGG and MetaCyc data.

  • Each KGML entry became an SBML Qualitative Species and each KGML relation became an SBML Transition.
  • KEGG relations describe non-molecular-reaction interactions including protein, enzyme, transcription-factor, gene, compound, and pathway links.
  • BioCarta information enriched KEGG relations by assigning subtypes based on activating or inhibiting control types.
  • Activation and expression became positive values, inhibition and repression became negative values, and other subtypes became unknown.
  • Relations containing both activating and inhibiting subtypes received the dual value, with SBO terms further specifying input semantics.
  • Genome-scale reconstructions used KEGG data for 2,630 organisms, with many models augmented by MetaCyc and identifiers reconciled through MNXref.
  • A shared default biomass reaction contained 20 amino acids, RNA and DNA nucleotide precursors, glycogen, and ATP, without organism-specific tailoring.The authors note anomalies such as glycogen inclusion in bacteria and plants, which can be manually curated.

The Systems Biology Graphical Notation

SBGN provides standardized graphical languages for process descriptions and activity flows. The project converted generated SBML models into automatically laid-out SBGN maps and stored them as SBGN-ML and PNG files.

  • SBGN Process Description represents chemical kinetics with molecular pools consumed and produced by reactions, whereas Activity Flow represents influences that inhibit or stimulate activities.
  • SBML Level 3 Core entries were translated into SBGN Process Description glyphs using SBO terms and stored layout positions as initial coordinates.
  • Entities with multiple connections were cloned so each displayed copy connected to a single element, producing structurally valid but initially incomplete maps.
  • New reaction positions were estimated from the availability and locations of reaction partners, placing unconnected reactions at the top of the map.
  • Final maps used a compact SBGN-conforming layout similar to the original KEGG layout and were exported as SBGN-ML and PNG files.
  • SBGN Activity Flow maps were generated from qualitative-model glyph locations and sizes, with glyph and arc types assigned from SBO terms.

Extension of BioModels Database to support the distribution of models

The distribution infrastructure was extended to handle, display, validate, browse, and release large collections of SBML models through BioModels Database.

  • Infrastructure: The infrastructure was upgraded for SBML Level 3 models using multiple SBML packages and a dedicated distribution branch.The underlying submission-to-release pipeline was also extended.
  • Model access: Model pages provide annotations, notes, downloadable SBML files, and graphical representations in PNG, SVG, and SBGN-ML formats.Users can also report issues through an online form.
  • Model ingestion: An automated submission tool reads models, performs checks, standardizes notes and annotations, extracts display information, and stores metadata and files.This supports consistent large-scale model ingestion.
  • Browsing and download: Users can browse models by organism and pathway or by model category, type, and organism.Three archives containing all models are available through EBI FTP servers.

FIGURE LEGENDS

The figure legends document workflows for converting pathway data into SBML and graphical models, enriching reconstructions, and characterizing the resulting model collection.

  • Workflow: Figure 1 traces extraction from pathway databases into SBML, addition of mathematical features and graphical descriptions, and distribution through BioModels Database.Examples of mathematical features include kinetic rate equations and flux bounds.
  • Graphical and encoded models: Figure 2 juxtaposes an SBGN Process Description map, a pathway cutout, and SBML fragments describing the displayed reactions.It links graphical pathway representation with its encoded reaction structures.
  • Kinetic enrichment: Figure 3 presents rate equations obtained from SABIO-RK for models from selected organisms.The figure focuses on kinetic information associated with reconstructed models.
  • Draft metabolic reconstruction: Figure 4 shows SuBliMinaL modules extracting and reconciling KEGG and MetaCyc data before merging draft reconstructions and adding media, transport, GPRs, and flux bounds.The resulting models are formatted for COBRA Toolbox analysis and represent the union of KEGG and MetaCyc information.
  • Collection overview: Figure 5 displays 2 630 genome-scale metabolic models on a phylogenetic tree, with colors for taxonomic groups and bars for four model statistics.The statistics are metabolic reactions, metabolites, makeable metabolites, and makeable biomass components.
  • Collection overview: Figure 6 zooms into the eukaryotic branch of Figure 5 and supports interactive zooming, searching, and statistic visualization through iTOL.It provides a more detailed view of the eukaryotic subset.
  • Model size: Figure 7 compares model classes by molecular species and mathematical relationships across whole-genome, qualitative, chemical-kinetic, curated, and non-curated models.The panels organize distributions by model class and BioModels Database branch.
  • Biological domains: Figure 8 groups model classes according to clustered Gene Ontology annotations representing biological domains.The displayed relative sizes are based on the models’ main GO annotations.
Loading 1307.7005v2…