Multimodal Digital Twin of the Pancreatic Beta-Cell

Developed directed graphical models for integrating models of the pancreatic beta cell at different length scales under a common statistical framework. Component sub-models included pharmacokinetic models of glucose–insulin dynamics, simplified Brownian diffusion models of molecular interaction between glucose and insulin vesicles, and ODE-based network models describing enzyme kinetics from glucose entry to Ca²⁺-mediated insulin vesicle exocytosis.

Built Bayes nets in PyMC3 and designed a lightweight API to automatically combine models into an overarching whole-cell model that can query arbitrary variable relationships — for example, how insulin vesicle translocation speed (a microscopic property) changes in response to glucose intake (a macroscopic property).

We named the approach “Bayesian metamodeling” and published a proof-of-concept paper as part of the Pancreatic Beta Cell Consortium. A tutorial is available in PyMC3.