Before building a complex digital twin, Takeda analyzed historical manufacturing data. A simple end-to-end model identified that changing set points for six parameters—within existing validated ranges—could boost yield by 35%, demonstrating massive value in data that companies already possess.
The acceptable ranges used in GMP manufacturing are typically defined by mathematical models built during process validation. Adopting a digital twin isn't introducing a foreign concept; it's simply deploying the same type of model in real-time with a human in the loop, rather than using a static, one-time calculation.
Instead of immediately building complex models for a single problematic unit operation, start with simple linear models that connect the entire process. This holistic view reveals which parameters truly impact the final drug substance, allowing you to focus resources on building complex models only where they are most critical.
Don't start by pitching a full-scale, real-time digital twin. Instead, use historical data to build an offline model and demonstrate concrete business value, such as a potential yield increase. This proven ROI makes it much easier to get executive backup and funding for the more expensive live implementation.
The regulatory pathway for deploying models in manufacturing is not an unknown frontier. It involves a credibility assessment for the model itself, following standards like ASME VNP 40, and established GAMP procedures for validating the data interfaces. This treats the model and its integration as distinct, manageable components with clear guidelines.