Quantifying and managing uncertainty in systems biology: Mechanistic and data-driven models

Abstract

Uncertainty poses a significant challenge to the reliability and interpretability of systems biology models. This review focuses on reducible epistemic uncertainty arising from incomplete data, measurement errors, or limited biological knowledge. We examine how this uncertainty affects both mechanistic models —such as dynamic kinetic and genome-scale metabolic models— and data-driven models, including neural networks trained on time-series data. Strategies for quantifying and mitigating uncertainty are reviewed, including profile likelihoods, Bayesian inference, ensemble modelling, optimal experimental design and active learning. Through illustrative case studies, we show how data limitations, model structure, and experimental design influence uncertainty propagation and model predictions. Finally, in our outlook, we highlight key research avenues to build more robust models, including hybrid frameworks combining mechanistic models with machine learning to improve interpretability and predictive performance, advances in inference methods and tools, or the definition of benchmarks to support reproducibility and method comparison.