Creating the Virtual Human Twin across scales

Virtual Human Twins that capture multiple scales offer unique possibilities to replace many animal experiments according to the 3R Principles. Digital twins of animal and other experiments can further contribute to reducing and refining animal experiments. At the Virtual Human Twin Core Unit, we develop computational models of small units such as cell distribution and cell fate in biopsies to microenvironments in tumors and organoids or even larger entities such as whole organs such as the heart. AI technologies can aid in personalizing and accelerating simulations with these computational models but also AI approaches can be enabled through large scale simulation studies in turn. Our expertise comprises technologies from computational image analysis, 3D tissue modeling, biomechanics and bioelectricity modeling and simulation, molecular biodiversity to biomedical data science.

A Virtual Human Twin is a personalized computational model of an individual patient. More general, a digital twin model mimics the structure and behavior of an individual or a technical object (e.g. a power plant) according to the definition of the AIAA Digital Engineering Integration Committee. Sometimes digital twin models are built for decision support in one critical timestep. While this is currently the most wide-spread use case in the medical field, digital twins in general are dynamically updated from the physical twin throughout their life-cycle, a vision for future developments also in the HealthTech field.

Digital Twinning
Digital twin workflow. A baseline model builds the basis for the digital twin. Often, it is a bottom-up mechanistic model informed by biophysical first principles and population-level knowledge. Anatomical and functional personalization is performed based on individual patient measurements (e.g. computed tomography and ECG). The parametrized digital twin can then be used in computational simulation to make predictions regarding personal risk or support decisions regarding optimal therapy for personalized medicine. The digital twin can be updated continuously when new measurements are available and refined by comparing predictions to real world outcomes.

Virtual Human Twins are used for individual risk prediction, decision support, and therapy planning. For example, different options for interventions can be evaluated in silico before deciding for the approach to be applied to the patient following the personalized medicine vision.

Computational modeling & simulation approaches
Computational modeling & simulation approaches. Representing an individual as well as possible with a computational digital twin model requires both anatomical and functional personalization (A). Once a digital twin is established, different interventions can be evaluated in silico to enable precision medicine (B). If anatomical and functional envelopes are continuous and limited to biologically relevant ranges, arbitrary numbers of new virtual subjects (digital chimeras) can be sampled from this space representative of a population but not of a specific individual in this population (C). With a whole cohort of digital twins or digital chimeras, in silico clinical trials can be conducted evaluating the effects of specific interventions consistently applied to the entire virtual study population (D).

One field in which Virtual Human Twins are quite advanced already is cardiology. We outline current applications and future visions in a chapter of the book Innovative Treatment Strategies for Clinical Electrophysiology. To exploit modeling & simulation technologies, we need to formulate the biological function in terms of mathematical equations as presented below for the example of cardiac electrophysiology across the cellular, tissue, organ and body scales.

Cardiac model
Hierarchy of multiscale cardiac electrophysiology models ranging from ion channels (A) via
integrated cell models (B) and tissue level models (C) to the body surface and electrocardiogram (D). The simulation system allows investigating what-if scenarios by changing input parameters of the model (top row) and analysing the effect on simulation outputs on numerous scales (bottom row).