SBML-SAT: a systems biology markup language (SBML) based sensitivity analysis tool

Zhike Zi, Yanan Zheng, Ann E. Rundell, Edda KlippBMC Bioinformatics 2008 · DOI

TL;DR

SBML-SAT is a new, freely available software tool that significantly enhances the ability to perform sensitivity analysis on systems biology models encoded in SBML. It addresses the limitations of existing tools by offering a comprehensive suite of local and global sensitivity analysis methods, support for models with discontinuous events, and an intuitive graphical interface, making complex model analysis more accessible.

What Problem Does This Paper Address?

Systems biology aims to understand complex biological systems by building mathematical models. These models are crucial for reproducing experimental data and predicting system behavior. However, these models are subject to various uncertainties, including experimental errors, incomplete data, and a lack of precise parameter knowledge. Sensitivity analysis is a vital technique to understand how these uncertainties, particularly in model parameters, affect the model's output and to identify the most influential parameters.

While many software tools exist for systems biology modeling, existing SBML-compatible tools often fall short in performing comprehensive sensitivity analyses. Specifically, they may:

  • Lack support for discontinuous events: Biological systems frequently exhibit events like cell division or signal removal at specific times, which many current tools cannot handle.
  • Offer limited global sensitivity analysis: Most tools focus on local sensitivity analysis, which examines parameter effects at a single point. Global analysis, which explores parameter variations across a wider range and their interactions, is often missing.
  • Provide inflexible output options: Users need to analyze various model outputs (e.g., steady-state, maximum, integrated responses), but existing tools may not offer sufficient flexibility in defining these.

This paper introduces SBML-SAT to fill these gaps, providing a powerful and versatile platform for sensitivity analysis of SBML models.

Methods

SBML-SAT is a MATLAB-based software package designed to perform simulation, steady-state analysis, robustness analysis, and both local and global sensitivity analyses for SBML models. Its key methodological contributions include:

  • Support for Discontinuous Events: Building on previous work, SBML-SAT can handle models with discontinuous events, including complex scenarios like cell division, broadening its applicability.
  • Comprehensive Sensitivity Analysis Methods:
    • Local Sensitivity Analysis: Implements standard methods using centered difference approximation for both unnormalized and normalized sensitivity coefficients. It allows for one-at-a-time (OAT) parameter perturbations.
  • Global Sensitivity Analysis: Offers four distinct methods:
  • Multi-Parametric Sensitivity Analysis (MPSA): Uses Latin Hypercube Sampling (LHS) to explore parameter space and measures sensitivity based on the Kolmogorov-Smirnov statistic.
  • Partial Rank Correlation Coefficient (PRCC): A rank-transformed linear regression analysis suitable for nonlinear, monotonic relationships.
  • SOBOL's Method: A variance-based method that apportions model output variance to individual parameters and their interactions.
  • Weighted Average of Local Sensitivities (WALS): Approximates global sensitivities by weighting local sensitivity indices based on model error.
  • Flexible Model Output Definition: Users can define model outputs as steady-state response, maximum response, integrated response, or time-dependent response, allowing for tailored analysis.
  • Steady State Analysis: Employs two methods: algebraic solution of ODEs for equilibrium (using KINSOL from SUNDIALS) and a quasi-steady-state method that simulates for a long time to find equilibrium.
  • Robustness Analysis: Quantifies how model outputs change with respect to total parameter variation (TPV) using a defined robustness metric.
  • Intuitive Graphical User Interface (GUI): Facilitates easy model loading, simulation, and analysis, with options to save settings and export results.

Datasets

The paper demonstrates SBML-SAT's capabilities using various SBML models from the BioModels Database. These include:

  • A MAPK cascade model (BioModels ID: BIOMD0000000010) for local sensitivity analysis.
  • A receptor trafficking network model for demonstrating global sensitivity, steady-state, and robustness analyses.
  • Other models such as a fission yeast cell cycle model (BIOMD0000000111), an NF-kB signaling pathway model (BIOMD0000000140), a T cell gene expression model (BIOMD0000000122), and a metabolic model (BIOMD0000000106) were used to showcase simulation capabilities across different model types, including those with discontinuous events.

No new benchmark datasets were introduced; rather, existing, publicly available models were utilized to validate the software's broad applicability.

Key Insights and Ideas

  • Bridging the Gap in SBML Analysis: SBML-SAT directly addresses a critical need for integrated and advanced sensitivity analysis tools within the SBML ecosystem, which is widely adopted for model sharing.
  • The Power of Global Sensitivity Analysis: The paper strongly advocates for global sensitivity analysis, highlighting its superiority over local methods for understanding parameter uncertainties in complex, nonlinear biological systems. The comparison of different global methods (MPSA, PRCC, SOBOL, WALS) demonstrates that while their specific metrics differ, they often yield similar rankings of parameter importance, increasing confidence in the findings.
  • Handling Biological Realities: The inclusion of support for discontinuous events is a significant practical advantage, as many biological processes are inherently dynamic and non-smooth.
  • Unified Analysis Platform: SBML-SAT consolidates simulation, steady-state, robustness, and multiple sensitivity analyses into a single, user-friendly interface, streamlining the workflow for systems biologists.

Key Results

  • SBML-SAT successfully simulates a variety of SBML models, including those with discontinuous events, demonstrating its broad applicability.
  • Local sensitivity analysis on the MAPK cascade model identified the initial concentration of MAPK as the most sensitive parameter for the integrated response.
  • Global sensitivity analyses on the receptor trafficking model revealed that parameters related to ligand-receptor complex formation (k₂ and k₃) are crucial for ligand concentration, while internalization and dephosphorylation rates (k₆, k₇, k₈) affect internalized complexes. MPSA also showed that the importance of a parameter can be time-dependent.
  • Robustness analysis indicated that unbound receptor concentrations are less robust to parameter variations than ligand-bound receptors.

Limitations and Future Work

  • No SBML Editor: SBML-SAT does not include an SBML model editor. Users must create their models using external tools like CellDesigner, COPASI, or SBMLeditor.
  • No Delay Differential Equation Support: Models involving delay differential equations are not directly supported. The authors suggest converting them to ODEs using linear chain transformation as a workaround.
  • Single Steady State: The steady-state analysis methods find only a single steady state to which the initial condition converges, potentially missing other stable or unstable steady states.
  • Future Work: The authors imply that future updates will be released on the software's website, suggesting ongoing development and potential for new features or improvements. The broad applicability and comprehensive nature of the tool suggest it will be a valuable resource for the systems biology community.

Summary updated

BibTeX
@article{zi2008sbml,
  author = {Zhike Zi and Yanan Zheng and Ann E. Rundell and Edda Klipp},
  title = {{SBML-SAT: a systems biology markup language (SBML) based sensitivity analysis tool}},
  journal = {BMC Bioinformatics},
  year = {2008},
  doi = {10.1186/1471-2105-9-342},
}
Figure