PEtab Select standardizes and automates model selection for systems biology

Model selection determines which processes to include when building mechanistic models of biological systems. Competing hypotheses lead to competing model structures, and comparing them requires repeated calibration (parameter estimation) and evaluation. Until now there was no widely adopted standard to specify model selection tasks in a compact, tool-agnostic way, which made large-scale or automated comparisons difficult to reproduce and reuse.

PEtab Select introduces a concise file-format standard and a supporting software package for specifying model selection problems and the associated calibration tasks. It extends the PEtab standard for parameter estimation so users can reuse existing workflows and tooling.

• Interoperable specification: A language-agnostic, standardized way to describe model alternatives, shared parameters, datasets, and calibration settings.

• Integration with established tools: Implementations work with COPASI, Data2Dynamics, PEtab.jl (Julia), and pyPESTO (Python), enabling users to run selection pipelines in environments they already use.
• Multiple selection criteria and methods: Built-in support for Akaike and Bayesian information criteria, with an extensible interface to add others. Model-space exploration strategies include brute-force enumeration, forward and backward selection, and more advanced flexible selection methods.

ReadTheDocs. The authors published code, data, and package versions for reproducing reported results on Zenodo. The package was developed by a community effort and integrated into multiple modeling frameworks to reach users across COPASI, Julia, MATLAB, and Python ecosystems.

• Scalability: The compact file format and available exploration methods make it feasible to handle very large model spaces that were previously impractical to evaluate exhaustively.
• Flexibility: Users can choose selection criteria and exploration strategies that match their computational budget and scientific question.

PEtab Select plugs into existing parameter-estimation pipelines. A typical workflow is: define model alternatives and calibration problems in PEtab Select format, use a supported backend (COPASI, Data2Dynamics, PEtab.jl, or pyPESTO) to perform parameter estimation or sampling for each candidate, and evaluate candidates with the chosen information criterion or metric. The package handles managing large sets of candidates and coordinating selection strategies.

Open access and community development

PEtab Select fills a gap in computational model workflows by providing a standard, tool-agnostic way to define, share, and run automated model selection problems at scale. It enables interoperability between major modeling tools and offers flexible exploration and evaluation options suitable for both small and very large model spaces.