[Paper Review] pyABC: Efficient and robust easy-to-use approximate Bayesian computation

Yannik Schälte, Emmanuel Klinger+2 more|arXiv (Cornell University)|Mar 24, 2022

Markov Chains and Monte Carlo Methods4 citations

TL;DR

This paper presents pyABC v0.12, an enhanced open-source Python framework for approximate Bayesian computation (ABC) that integrates advanced algorithms for robust, efficient, and automated likelihood-free inference. It introduces adaptive distance scaling, outlier robustness, informative summary statistics via regression, optimal transport distances, and threshold optimization to improve posterior approximation across diverse models and data types with minimal user tuning.

ABSTRACT

The Python package pyABC provides a framework for approximate Bayesian computation (ABC), a likelihood-free parameter inference method popular in many research areas. At its core, it implements a sequential Monte-Carlo (SMC) scheme, with various algorithms to adapt to the problem structure and automatically tune hyperparameters. To scale to computationally expensive problems, it provides efficient parallelization strategies for multi-core and distributed systems. The package is highly modular and designed to be easily usable. In this major update to pyABC, we implement several advanced algorithms that facilitate efficient and robust inference on a wide range of data and model types. In particular, we implement algorithms to account for noise, to adaptively scale-normalize distance metrics, to robustly handle data outliers, to elucidate informative data points via regression models, to circumvent summary statistics via optimal transport based distances, and to avoid local optima in acceptance threshold sequences by predicting acceptance rate curves. Further, we provide, besides previously existing support of Python and R, interfaces in particular to the Julia language, the COPASI simulator, and the PEtab standard.

Motivation & Objective

• To develop a user-friendly, scalable, and robust ABC framework for likelihood-free parameter inference in complex models.
• To reduce manual hyperparameter tuning by implementing self-tuning algorithms for transition kernels, population sizes, and acceptance thresholds.
• To improve inference accuracy and efficiency on noisy, outlier-affected, or high-dimensional data through adaptive distance metrics and informative summary statistics.
• To support heterogeneous modeling environments by integrating with Julia, R, COPASI, and the PEtab standard.
• To enable scalable inference via dynamic parallelization on multi-core and distributed systems.

Proposed method

• Implements a sequential Monte Carlo (ABC-SMC) scheme with adaptive transition kernels and self-tuned population sizes for improved convergence.
• Introduces adaptive scaling of distance metrics using iterative reweighting to handle data points with varying scales.
• Applies robust outlier detection and suppression via a modified version of Prangle (2017) to reduce the impact of measurement errors.
• Uses inverse regression models (e.g., linear, neural networks, Gaussian processes) to generate informative, data-driven summary statistics.
• Employs optimal transport-based distances to bypass the need for manual summary statistics and improve robustness to data structure.
• Predicts acceptance rate curves to avoid local optima in threshold sequences, enabling more stable and efficient SMC adaptation.

Experimental results

Research questions

• RQ1 How can ABC inference be made more robust to data noise and outliers without manual intervention?
• RQ2 Can adaptive distance scaling improve posterior approximation in models with heterogeneous data scales?
• RQ3 To what extent can regression-based summary statistics enhance inference efficiency and accuracy in ABC?
• RQ4 How can optimal transport distances replace summary statistics while preserving inference quality?
• RQ5 Can predictive modeling of acceptance rate curves prevent suboptimal threshold sequences in ABC-SMC?

Key findings

• The integration of adaptive distance scaling and outlier robustness significantly improves posterior estimation on real-world data with measurement errors.
• Regression-based summary statistics lead to higher-order posterior approximations and reduced dependence on user-defined summary statistics.
• Optimal transport-based distances enable effective inference without summary statistics, particularly beneficial for complex or high-dimensional data.
• Predictive modeling of acceptance rate curves prevents local optima in threshold sequences, improving convergence stability.
• Dynamic parallelization strategies in pyABC enable efficient scaling across multi-core and distributed computing environments.
• The framework supports seamless interoperability with Julia, R, COPASI, and the PEtab standard, enhancing usability across systems biology and computational modeling workflows.