Predictive Freeze-Drying Model Designed for Continuous Processing

Interest in intensified manufacturing is driving innovation in downstream technologies, according to the team behind a new predictive freeze-drying model designed with continuous processes in mind. Lyophilization, or freeze drying, is used to stabilize protein drugs by removing residual water through a series of cooling and drying steps. At present, it is very common—according to one study, half of all approved large molecule drugs are freeze-dried.

Whether lyophilization remains the stabilization technique of choice will depend on how well it can be adapted for continuous production methods, according to Prakitr Srisuma, a doctoral researcher at the Massachusetts Institute of Technology (MIT). “In general, continuous manufacturing has several benefits over batch or semi-batch operation, and that is also true for lyophilization. When a process is operated continuously, downtime is minimized, resulting in a higher production capacity. In addition, since a continuous process is not frequently interrupted like in batch or semi-batch processes, more consistent process and quality control can be obtained,” Srisuma tells GEN.

With this in mind, Srisuma and colleagues have developed a computer model that can predict how freeze-drying technologies will perform when used in continuous mode. “Our mechanistic model for end-to-end continuous lyophilization captures all three key steps in the process, namely freezing, primary drying, and secondary drying. Practitioners can use this model to study the entire process as a whole or study each part of the process individually,” he says. The model, described in a recent paper, tracks key process parameters—temperature, moisture content, phase composition, and freezing time—and was designed to balance accuracy with computational performance. Srisuma says, “Each simulation takes less than one second on a normal laptop but provides reasonably accurate results. Hence, users can employ our model to improve their continuous lyophilization processes in several ways: from design optimization, process monitoring/state estimation, and model-based control.”

In a separate study, Srisuma and colleagues used the model to identify the optimal operating conditions for a biopharmaceutical manufacturing process and were able to reduce drying time by several hours. The model, which is available on the code sharing website GitHub, is designed to run on the programming languages MATLAB and Julia. In the future, Srisuma believes the model could play a role in process control in combination with automation and artificial intelligence (AI) systems. “Automation is, of course, important for process operation, monitoring, and control of continuous lyophilization. It can play a role in ensuring the process can be operated smoothly and continuously with minimal human intervention. AI machine learning could be used to improve the accuracy of the models or complement mechanistic modeling where the underlying physics are not well understood,” he said.