Personalizing computational models to construct medical digital twins.

Abstract

Digital twin technology, originally developed for engineering, is being adapted to biomedicine and healthcare. A key challenge in this process is dynamically calibrating computational models to individual patients using data collected over time. This calibration is vital for improving model-based predictions and enabling personalized medicine. Biomedical models are often complex, incorporating multiple scales of biology and both stochastic and spatially heterogeneous elements. Agent-based models, which simulate autonomous agents, such as cells, are commonly used to capture how local interactions affect system-level behaviour. However, no standard personalization methods exist for these models. The main challenge is bridging the gap between clinically measurable macrostates (e.g. blood pressure and heart rate) and the detailed microstate data (e.g. cellular processes) needed to run the model. In this article, we propose an algorithm that applies the ensemble Kalman filter, a classic data-assimilation technique, at the macrostate level. We then link the Kalman update at the macrostate to corresponding updates at the microstate level, ensuring that the resulting microstates are compatible with the desired macrostates and consistent with the model's dynamics. This approach improves the personalization of complex biomedical models and enhances model-based forecasts for individual patients.