Hybrid Modelling Frameworks in Bioprocessing: Current Practices and Future Directions

Abstract

Bioprocessing encompasses the development, control, and optimization of biologically driven systems, which are inherently nonlinear, multiscale, and often poorly characterized. Mechanistic models based on first principles offer deep insight into system behaviour but are frequently limited by structural rigidity, high parameterization demands, and difficulty in capturing biological variability. In contrast, data-driven models, comprising artificial intelligence (AI) and machine learning (ML) methods, can offer strong predictive capabilities but lack physical interpretability. Hybrid modelling frameworks - integrating mechanistic and data-driven approaches are emerging as powerful tools to address these limitations, offering improved accuracy, adaptability, and interpretability. In this article, we offer a comprehensive overview of hybrid modelling strategies applied across upstream and downstream bioprocesses. We examine classical and modern modelling paradigms, highlighting the motivations for hybridization and the structural architectures commonly used. A comparative framework summarizes model types, evaluation metrics, and outcomes, providing practical insights for researchers and practitioners. The paper also highlights emerging trends such as the incorporation of transfer learning, explainable AI, and self-adaptive systems within hybrid models. Challenges, including data scarcity, model generalization, and regulatory considerations, are discussed, alongside strategies for enhancing interdisciplinary collaboration and standardization. The article concludes by identifying future research directions that advocate for standardized, interpretable, and robust hybrid modelling pipelines, with the potential to drive intelligent bioprocessing and accelerate the transition toward digitalized biomanufacturing ecosystems.