Modeling the human body
Human physiology experiments are generally slow, expensive or unsafe. A reliable physiological model makes it possible to explore those questions in-silico first . Thereby, reducing costs and significantly speeding up research and development processes.
Existing solutions
Today's models of human physiology are mostly a collection of statistical models of isolated mechanisms. The models miss variables that matter and only work within the conditions considered during development. This is hardly surprising: many variables cannot be measured, data from extreme conditions is scarce, and every physiological measurement is inherently extremely noisy. Together, this makes it nearly impossible for a purely statistical approach to predict human physiology outside familiar conditions.
Consequences
-
Low generalizability
Limited data combined with purely statistical models leads to large errors outside the training context. Research shows a 1.7–2.8x; increase in error with only a slight shift in exercise intensity (Psikuta et al., 2011).
-
New scenarios cannot be modelled
Conditions that were never measured cannot be predicted. I.e. a model fitted on standard conditions has no way to predict physiological responses during, for example, severe blood loss.
-
Limited understanding
Statistical models can predict, but they do not explain. It remains unclear why a result occurs or what would change it. Questions like Why did the model predict this? and What should we change to improve the outcome? remain unanswered.
Our approach
Instead of learning patterns from limited data, we build a mechanistic model: a model based on the physical and physiological laws that govern the body, such as pressure laws, the laws of heat transfer, and thermoregulatory control pathways. This approach combines three ingredients:
Mechanistic models
Physical and physiological laws hold in every condition, so the model keeps working beyond the data it was built on.
Domain knowledge
Decades of physiological research describe how the body's systems interact, so we do not need to start from scratch.
Machine learning
Where mechanisms are still unknown or unmeasurable, data-driven methods fill the gaps within a physically consistent model.
Roadmap
- Sep 2024 Done Thesis on predicting body temperatures
- Oct 2026 Done Funding secured
- Nov 2026 Next Full-time model development begins
- Feb 2027 V0.5. Mechanistic physiology model
- May 2027 V1.0. Integrated whole-body model
- Nov 2027 V1.5. Model refined against published literature
- May 2028 V2.0. Model refined with specific experimental measurements
- Nov 2028 Model ready for use in research and development
- 2030 Physiological simulation platform for human health, safety and performance
Follow our progress
Get an occasional update when we reach a milestone on the roadmap.