What WW2 medicine taught us about the brain
Much of what neuroscience knows about how the brain divides its work came from the Second World War. Soviet neuropsychologist Alexander Luria spent the war treating soldiers with penetrating head wounds, and by matching the location of each injury to what the patient could and could not do afterward, he built a map of brain function that founded modern neuropsychology.
Many people exposed to the same extreme condition, recorded over time, reveal mechanisms that ordinary life does not show.
Spaceflight changes the body in weeks rather than decades, and every crew member who flies adds to a record that, gathered in one place, can become a model of how the body adapts.
In space, we age much faster
In NASA's Twins Study, a year in orbit changed Scott Kelly's telomeres, the activity of thousands of genes, and left radiation damage in his chromosomes. Some changes had not reversed six months after landing.
In orbit the body shows signs of ageing far faster than on Earth, and we still do not know why. The data comes from very few people. A model can combine every crew member's measurements and test what drives these changes.
Cells can return to a fetal-like state
When tissue heals, adult cells briefly step back into a fetal-like state and rebuild from there. A short burst of the Yamanaka factors triggers the same step without an injury, and a 2023 study in Science Advances traced both routes to prostaglandin synthesis.
Prostaglandin pathways are already familiar drug targets. Whether spaceflight, with its rapid loss and recovery of tissue, shifts this repair program is an open question that a model built on crew data could test.
Drugs and treatments that started in a model
A model does not replace a trial but it narrows millions of options to the few worth testing
Halicin
A deep learning model at MIT screened over 100 million molecules and found an antibiotic that kills resistant bacteria. Cell, 2020.
Abaucin
The same approach found a narrow-spectrum antibiotic against Acinetobacter baumannii, one of the hardest hospital infections to treat. Nature Chemical Biology, 2023.
Rentosertib
Insilico Medicine used AI to pick both the target and the molecule. It is the first drug discovered this way to report Phase 2a results. Nature Medicine, 2025.
A simulated patient population
The FDA accepted the UVA/Padova simulator of type 1 diabetes in place of animal testing for insulin dosing algorithms. A virtual patient did the work of the first experiment. J Diabetes Sci Technol, 2009.
From one experiment to the whole literature
The usual way to check a model of the body is one experiment at a time: run a study, run the same study in the model, compare. We are building the simulation the other way round. Agents gather every published spaceflight and analog result that applies, test the model against each one, and report which results the model reproduces and which it cannot.
The gaps are the useful part. Each result the model cannot explain points to a mechanism we do not yet understand, and that becomes the next experiment, on the ground or in flight.
The agents work through a shared lab notebook rather than a chat. Every assumption, run and result is written down where the team can read, audit and correct it, so the model stays something scientists can trust.