[Paper Review] pyPESTO: A modular and scalable tool for parameter estimation for dynamic models

Yannik Schälte, Fabian Fröhlich+15 more|arXiv (Cornell University)|May 2, 2023

Gene Regulatory Network AnalysisBiochemistry, Genetics and Molecular Biology19 references3 citations

TL;DR

pyPESTO is a modular, open-source Python framework for scalable parameter estimation in dynamic models, particularly ordinary differential equation (ODE) systems. It unifies optimization, uncertainty quantification, and visualization through a standardized pipeline, supporting both local and global solvers, Bayesian and frequentist inference, and integration with popular tools like AMICI and PyMC.

ABSTRACT

Mechanistic models are important tools to describe and understand biological processes. However, they typically rely on unknown parameters, the estimation of which can be challenging for large and complex systems. We present pyPESTO, a modular framework for systematic parameter estimation, with scalable algorithms for optimization and uncertainty quantification. While tailored to ordinary differential equation problems, pyPESTO is broadly applicable to black-box parameter estimation problems. Besides own implementations, it provides a unified interface to various popular simulation and inference methods. pyPESTO is implemented in Python, open-source under a 3-Clause BSD license. Code and documentation are available on GitHub (https://github.com/icb-dcm/pypesto).

Motivation & Objective

• To address the challenge of estimating unknown parameters in complex, dynamic biological models, especially those governed by ordinary differential equations (ODEs).
• To provide a unified, extensible framework that supports diverse inference methods for parameter estimation in systems biology.
• To enable scalable and efficient parameter estimation through modular design, parallelization, and interoperability with established simulation and inference tools.
• To support both frequentist and Bayesian uncertainty quantification methods, including profile likelihood and Markov chain Monte Carlo sampling.
• To streamline the entire parameter estimation workflow—from problem definition to result analysis—via a consistent, user-friendly interface.

Proposed method

• Uses the PEtab standard for specifying parameter estimation problems, enabling interoperability with existing model repositories and tools.
• Provides a unified interface to multiple optimization libraries (e.g., Ipopt, Dlib, SciPy, NLopt, PySwarms, pycma) and supports both local and global optimization strategies.
• Supports derivative computation via user-defined functions or adaptive finite differences, improving convergence speed and accuracy.
• Implements hierarchical optimization for relative data and noise parameters, enhancing robustness in high-dimensional problems.
• Integrates Bayesian sampling methods (adaptive Metropolis, adaptive parallel tempering) and supports gradient-based samplers via interfaces to Emcee, PyMC, and Dynesty.
• Offers HDF5-based result storage and visualization tools, enabling efficient data handling and reproducibility.

Experimental results

Research questions

• RQ1 How can a modular and extensible framework improve the reproducibility and scalability of parameter estimation in dynamic biological models?
• RQ2 What is the performance of pyPESTO’s multi-start globalization strategy in solving non-convex, multi-modal parameter estimation problems common in systems biology?
• RQ3 How effectively can pyPESTO unify diverse optimization and sampling tools into a single, consistent interface for ODE-based parameter estimation?
• RQ4 To what extent does pyPESTO’s support for both frequentist (profile likelihood) and Bayesian (profile posterior) uncertainty quantification enhance model reliability?
• RQ5 How well does pyPESTO scale across different computational infrastructures, from desktops to high-performance computing clusters?

Key findings

• pyPESTO enables systematic parameter estimation across a wide range of dynamic models, with support for both ODE-based and black-box parameter estimation problems.
• The framework successfully integrates with established tools such as AMICI for ODE simulation and PyMC for Bayesian inference, enhancing usability and interoperability.
• pyPESTO’s multi-start globalization strategy effectively navigates complex, multi-modal likelihood landscapes, improving convergence to global optima.
• The implementation of both frequentist and Bayesian uncertainty quantification methods—via profile likelihood and MCMC sampling—provides robust confidence assessment for estimated parameters.
• The use of HDF5-based result storage enables efficient, compact, and reproducible handling of large-scale estimation results.
• pyPESTO has already been adopted in at least 12 publications and is actively maintained by a core team across three institutions, indicating strong research utility and community adoption.