A neural ordinary differential equation model for predicting the growth of Chinese Hamster Ovary cell in a bioreactor system

Abstract

Chinese hamster ovary (CHO) cells play an important role in the biopharmaceutical industry, but their production efficiency requires enhancement to meet the growing market demands. Artificial intelligence (AI) has been a potent tool for modeling bioprocesses to support biopharmaceutical manufacturing. However, existing AI models do not adapt well to process data collected at irregular time intervals and have limited capability to scale up and down to incorporate various process parameters. To address the limitations, this study develops a novel neural ordinary differential equation (ODE) model for predicting key variables such as viable cell concentration, glucose concentration, lactate concentration, pH, and dissolved oxygen in a CHO cell bioreactor. Validated through extensive bioreactor experiments, the neural ODE model shows a better accuracy compared to the benchmark models, which include a conventional mechanistic model and a hybrid model. Additionally, the neural ODE model incorporated essential process variables that were not considered in the previous models. It successfully extrapolates to predict unknown dynamics at different initial conditions, which showcases robust adaptability. Moreover, the model provides useful insights into the relationship among variables, highlighting its potential for bioprocess modeling.