Digital Design for Advanced Biotherapeutics: A Model-enabled Framework for the Development and Optimisation of Biopharmaceutical Cell Culture Processes

Driven by scientific innovation and apposite regulatory guidance, the biopharmaceutical industry continues to evolve and adapt to meet the growing needs of patient populations worldwide. Prospective pipelines for investigative biotherapeutic candidates continue to expand, while the advent of biosimilars and further promise of advanced therapy medicinal products place greater economic uncertainty and commercial pressure on the development of new therapeutic entities. Clinical and commercial success is largely contingent upon the timely and efficient development of quality biotherapeutic candidates. Of equal importance, the development of a robust manufacturing strategy is paramount to ensure the therapeutic candidate can be reliably produced with the desired quality profile to consistently deliver the demonstrated clinical performance. Implementation of the Quality by Design (QbD) framework greatly contributes towards assurance of product quality with the systematic identification of critical quality attributes (CQAs), and subsequent linking of process inputs to CQAs though critical process parameters (CPPs) and critical material attributes (CMAs). By closely integrating quality attributes within the manufacturing process, product quality may be designed into the process during development, providing greater confidence and assurance in the established manufacturing strategy. To effectively link product quality with the manufacturing process, statistical and mathematical models provide an effective approach for quantitatively identifying the relationship between CPPs, CMAs, and product CQAs. While limitations have been identified across both statistical and mechanistic modelling approaches, novel hybrid modelling strategies, which combine statistical malleability with mechanistic certainty, are expected to advance process development programs through reduced model development effort, improved predictive performance, extended extrapolative capability, and capacity to capture the influence of process inputs and state variables on cell culture dynamics. Hybrid models thus have the potential to accelerate bioprocess development while ensuring the most efficient use of experimental resources during the investigative campaign. This research delivers a systematic approach to the development, evaluation, and application of hybrid dynamic models for fed-batch CHO cell culture processes, with a particular emphasis on understanding the relationship between model structure, data processing workflows, and predictive performance. By leveraging both high-throughput technology and in-silico strategies for data generation, this work quantifies the impact of data quality and quantity on model reliability and predictive capacity. Comparative analysis of Gaussian Process Regression and Artificial Neural Networks provides insights into their respective strengths and limitations for estimating cell-specific rates within a hybrid framework. The developed models are applied for computational optimisation, with a focus on identifying optimal extracellular pH trajectories, further demonstrating their practical utility in enhancing upstream process development and supporting the deployment of model-informed strategies, aligned with QbD principles. The presented research contributes a structured and scalable approach, that not only strengthens the methodological foundation of mammalian cell culture modelling, but also provides a flexible framework to foster innovation in bioprocess development and manufacturing operations.