(371al) Model Predictive Control in Continuous Biomanufacturing Utilizing In-Line Process Analytical Tools

The industrialization of integrated continuous manufacturing is of growing interest in the biopharmaceutical industry due to the increased process intensification and manufacturing flexibility that can be achieved within a reduced operational footprint. To fully achieve the benefits and realize the advantages relative to traditional batch-based methods, process analytical technologies (PAT) are required to enable real-time monitoring of critical process parameters (CPP), critical quality attributes (CQA), and deployment of effective process control strategies.

Downstream purification involves various unit operations that purify, polish and concentrate the drug substance (API). Tangential flow filtration (TFF) is commonly used for concentrating the protein product. Operating in a single-pass mode enables the feasibility of integrating TFF into a continuous purification process. Controlling and maintaining a target protein concentration while operating at a target transmembrane pressure, is a function of the incoming feed concentration, flowrate, and membrane area. In a traditional TFF unit operation, protein concentration is periodically measured offline and transmembrane pressure is controlled through two P-I-D loops: one on a back-pressure valve that opens or closes to modulate target pressure and another on a pump that regulates feed flow into the system. While PID loops can effectively control each input to a single respective output, the complexity of a single-pass TFF (SPTFF) operation consisting of multiple interrelated variables drives the need to introduce a control strategy that can account for parameter interactions.

This work describes the implementation of Model Predictive Control in continuous SPTFF utilizing a custom designed PAT tool that measures protein concentration in real-time. Considerations for enabling continuous operation involve demonstrated robustness to variations in incoming product profiles, while reaching target end points for the process. The controlled variables in SPTFF operate with significant differences in response times while adjusting to changes in process parameters. The model predictive controller is hence trained and tuned appropriately to account for such system dynamics and utilized to control concentration and TMP to setpoints across disturbances, facilitated through real-time information relay from inline PAT and process sensors.