Mathematical models that describe a physical system and are updated in real-time are often called digital twins. We describe our digital twin efforts in two areas: (1) automated insulin delivery for type 1 diabetes, and (2) biotherapeutics manufacturing.
Automated insulin delivery systems are composed of a continuous glucose monitor (CGM), a continuous insulin infusion pump, and a controller (often on a mobile phone) that changes the insulin pump delivery rates based on the CGM signals. Meals represent a major challenge because a large insulin bolus is needed to compensate for the carbohydrates in a meal. Ideally an individual would estimate the carbohydrate content of their meal and directly bolus the insulin at mealtime. Unfortunately, people often forget to do this, causing a large ""post prandial"" glucose excursion. We provide clinical results based on a multiple model probabilistic controller (MMPPC) that anticipates meals and their future effects but also considers the risk of over-delivering insulin and causing hypoglycemia.
Bioreactors are used to produce therapeutics, including monoclonal antibodies or autologous CAR-T. For CAR-T we discuss the use of digital twins at multiple scales. One is the expansion bioreactor that grows the cells by several orders of magnitude before infusion. Another represents the patient and their health state throughout the treatment. We present a conceptual framework and data and analysis tools to improve treatment decision-making.