(234c) Generalizing Cell Culture Models from Fed-Batch to Continuous Using Metabolic Knowledge

Continuous biomanufacturing has the potential to significantly enhance the yield, productivity, and consistency of therapeutic protein production. However, the transition of cell lines which are typically optimized for fed-batch operation to continuous mode presents a challenge, particularly in selecting a priori satisfactory continuous operating parameters.

Current methodologies rely on extensive experimentation, which is time- and cost-intensive, even when aided by design of experiments. This motivates the need for model-based methods that can reduce the experimental load by predicting performance, based on existing data and prior knowledge, leveraging experimentation only for final validation. Existing models, however, are overparameterized to experimental conditions, thus not generalizable across cell lines, feeding strategies and other operating parameters. It is possible to increase a model’s generality, but this requires more data and thus more experimentation, which ultimately incurs the same costs. Neither of these approaches are satisfactory as they do not address the underlying problem of requiring extensive and costly experimentation.

We propose overcoming the need for extensive experimentation by building more general reactor models through embedding first-principles knowledge. By incorporating cell metabolism in a nonparametric model of the production reactor, we are able to predict key process parameters like metabolite concentrations without requiring kinetic parametrisation. By limiting ourselves to the core metabolic reactions common across different mammalian cells, our novel multi-scale formulation offers increased predictive capability and generality. Additionally, by introducing mode and scale-specific constraints, our formulation can predict across different modes of operation without requiring kinetic parameter estimations, thereby reducing the need for extensive experimental data.

We present an evaluation of our model by comparing its predictions against both fed-batch and continuous industrial cell culture data. This novel formulation can be used to estimate performance for transitioning from fed-batch to continuous, by incorporating mode-specific constraints. The applicability of this formulation is extended to control, namely soft sensing of hard-to-measure metabolites and to estimate when to finish the culture optimally, further demonstrating the versatility of the proposed model.