Automated assembly of hybrid dynamic models for CHO cell culture processes

The emergent realisation of Industry 4.0 principles across biomanufacturing, through recent endeavours, will markedly enhance the development and manufacture of modern therapeutics. Through implementation of digital process models, a greater understanding of the intricate relationship between product quality attributes and manufacturing process performance may be established. While contributing towards accelerated process development, representative process models enable advanced optimisation of process parameters, thus having a tangible impact on the assurance of product quality and manufacturing robustness. Hybrid approaches, which couple mechanistic interpretability with statistical data-fitting, are posed to broaden the value and utility of digital models. To augment the advancement in modelling techniques and high-throughput technology, there is a growing requirement for automated approaches towards data processing and model assembly. In this study, a novel strategy is proposed, which leverages saturation and sigmoidal relationships, along with an underlying material balance framework, for the automated assembly of hybrid dynamic models of cell growth. The proposed hybrid model is compared against an equivalent mechanistic model based on Monod expressions. While both models achieve a reasonable fit against experimental data, the hybrid model demonstrates superior predictive performance. Development of automated hybrid models, as demonstrated in this study, may greatly accelerate process digitalisation across biopharmaceutical manufacture.