In physical systems, order emerges because matter and energy are constrained by governing laws. Biological systems inherit this property.
Every protein that folds, every molecule that binds, every cell that changes state and every disease that progresses does so within the laws of the physics.
Our technology is based on a simple premise:
If the physical conditions in which biology occurs can be replicated in silico, we can observe how a phenomenon unfolds through time.
Micro World Models have these physical conditions encoded into them.
Rather than inferring patterns from training data, these physical conditions act as boundary constraints.
In 2024, we designed a bioreactor to produce biological consumables for deep-space missions. To do this, we needed to understand how physics impacts biology dynamically.
This research led to the development of a novel biophysics framework used as the foundation of our technology.
Where LLM foundation models are trained on millions and millions of static "snapshots" in time (crystallized proteins, omics data, biopsies etc), Micro World Models are first principles and allow the dynamic computation of biology without training data.
We use the same biophysics framework across our space research and disease models, configuring the relevant physical constraints for the question being asked.
For example, the same protein can exist in the brain, heart, liver and skin but operate under different conditions, including pressure, chemical composition and electrical activity. We first define the environment and boundary conditions so the model reflects the biology we want to investigate.
Once the constraints are defined, we introduce an input, such as a gene sequence, which triggers our proprietary recursive mathematical cascade.
At each step, the algorithm applies the physical constraints to eliminate impossible states. Each resolved state changes the conditions for what comes next.
This allows us to model biology as it unfolds and retaining each successive state change creates a complete computational history.
This process creates a 4D computation of biology, where complex behaviour emerges from physical constraints and molecular interactions over time.
Micro World Models retain the full computational history, revealing how proteins form, interact and change (and where disease alters that process). Each computation expands our library of 4D proteins and molecular interactions, creating a growing proprietary resource for investigating disease mechanisms and opportunities for treatment.
So, if we don't use training data, what are inputs and outputs of our models?