A digital twin of the brain depends on complete measurements and continuous updates
Nature Reviews Electrical Engineering review: personal measurements determine the accuracy of a digital twin of the brain
On 10 August, Nature Reviews Electrical Engineering published a review of digital twins of the brain. The authors propose evaluating such a model by the resolution, completeness, and update frequency of data from the specific person.
A digital twin of the brain is a model of one person. Its structure and function are compared with measurements from that person's brain, and new data should refine the model. The open problem, as the authors frame it, is to describe one living brain.
The review states this criterion as follows:
“The achievable accuracy of a digital twin is determined by the resolution, completeness, and update frequency of these measurements, not by the number of neurons.”
The number of neurons indicates the scale of the computation. The accuracy of a model for a specific person depends on the detail and completeness of that person's data, how often the data are updated, and whether the model's calculations are validated against them.
A previous report on the state of brain emulation noted that no organism has yet had electrical signals recorded from more than 95% of its neurons across more than 95% of its brain volume. The new review proposes assessing this limit through the properties of the measurements.
For the human brain, the authors propose combining data at three scales: electron microscopy of small tissue samples, statistical estimates of connections between neurons, and MRI, which provides an image of the entire organ. This approach allows observations from small regions to be compared with measurements of the brain as a whole.
This type of validation is already being tested in the Eon research program using small neuronal cultures. Researchers first record the activity of one culture, then determine its structure and test whether the model predicts the recorded signals. Any error indicates which data the model lacks.
In the December 2024 Digital Brain study, the authors built a model from personal MRI data and biological constraints, with up to 86 billion neurons and 47.8 trillion synapses. They compared it with resting-state functional MRI data, which show changes in blood flow associated with brain activity, and tested the model's response to visual signals.
For a model of one person, the authors propose evaluating the quality of its connection to that person's data: exactly what has been measured, how those measurements are used to validate the calculations, and how the model is updated with new data.