In silico approaches towards automated biomanufacturing
The goal of manufacturing is to control and maintain a commercial process capable of consistently producing a drug of the highest quality and safety. As part of manufacturing science and technology (MSAT), mechanistic modeling can be used to support several acitivities in pharmaceutical manufacturing, including process transfer activities, in silico troubleshooting, process analytical technologies and automated biomanufacturing.
The manufacturing processes of pharmaceutical proteins require scale-up and process transfer activities at several stages of the product lifecycle to meet clinical and commercial product demands. Scale-up and process transfer activities are, for example, needed to scale the process from lab scale to pilot scale and finally up to full commercial scale. To guarantee product quality throughout the product lifecycle, it must be ensured that the process performance remains unaffected by differences in scale and that process knowledge based on scale-down model data can be used to claim robust operating ranges for all process scales. Scale independence is also important to avoid facility-fit challenges during the transfer to the full-scale commercial system.
Mechanistic process models can support process transfer activities as they are applicable across all scales. They describe the behavior of the process as a function of system parameters like the dimension of the chromatography column or its packing quality. Differences in scale can therefore be mimicked by the mechanistic model and how these differences affect process behavior can be investigated. By doing so, they can also be used early on to investigate whether a process fits into an existing production facility. This can be important for flexible production, multi-product facilities or a smooth transition to continuous processing.
Despite all the precautions taken during product development and manufacturing, the risk of a deviation in the manufacturing process can never be fully eliminated. Variations in the process, such as the aging of the chromatography column, can lead to a change in product quality and eventually the loss of a manufacturing batch. Identifying the root cause for variations in product quality is not always straightforward, as process readouts like UV and conductivity traces provide no direct evidence.
As mechanistic models provide a causal relationship between process input and process output (product quality), they cannot only predict how a change in the process input affects its output but can also inversely identify root causes for an observed process outcome. By systematically changing process parameters and raw material properties in silico, process conditions can be identified that describe observed deviations during manufacturing. Thereby, the model-based approach does not just identify the root cause of the deviation but also provides information on the quantity of the deviation. As an in silico root cause investigation is significantly faster than searching relevant historic data sets or starting an experimental investigation, mechanistic models can help to identify process deviations more quickly and reduce the risk of batch failure. Along with drugs which are more difficult to produce and a growing complexity of manufacturing processes, root cause investigations are becoming an increasingly high dimensional and non-linear problem.
Multiple process parameters and material attributes jointly contribute to and modulate a process deviation. In silico troubleshooting is the method of choice to identify those high dimensional problems and to resume production without delay.
In the past, product quality in manufacturing was ensured by release testing, operated separately from the actual manufacturing process and at a different time and place. Today, real-time analytics on the manufacturing floor are gaining more importance than separated analytics.
Although testing the end product is an integral part of today’s quality control strategies, processes are increasingly designed towards Quality-by-Design (QbD) to ensure the highest quality and process robustness from the early process design onwards. There is a body of documents from regulatory authorities framing the use of QbD elements which promote a science-based and data-driven approach through process analytical technologies (PAT).
PAT is based on systems for analyzing and controlling manufacturing through the real-time measurement of quality attributes and performance indicators with the goal of improving process understanding and ensuring final product quality. Mechanistic process models facilitate the most advanced soft sensors, essential for the implementation of PAT in manufacturing.
Biopharmaceutical processes are increasingly being automated to prevent operator errors and eliminate potential sources of variability. Efficient process automation requires robust control strategies. Since pharmaceutical processes are a sequence of individual unit operations, the control strategy should not be limited to a single unit operation but should consider interactions between unit operations as well. Within an automated downstream process, mechanistic models can be the key enabler for a systematic use of model predictive control (MPC). Unlike a traditional controller, which aims to minimize the deviation of a process variable from its set point, MPC uses dynamic models such as mechanistic chromatography models to predict the future behavior of the process as a function of the process variables observed in the past. By combining mechanistic models for each unit operation, they can help to better understand interactions between unit operations, enable process control in a feed forward manner, and to implement unified process control.