Patrick Mineault proposes training AI on cell culture data and transplant outcomes to accelerate neuron replacement

Patrick Mineault proposes training AI on cell culture data and transplant outcomes to accelerate neuron replacement

In his review dated 30 October 2026, neuroscientist Patrick Mineault traced the development of neuron replacement from neural tissue transplants in the 1980s to current cell therapies. On 6 March, Japan granted conditional, time-limited approval to one such therapy for motor symptoms of Parkinson’s disease inadequately controlled by medication.

Parkinson’s disease causes the loss of neurons that produce dopamine, a substance the brain uses to regulate movement. Immature cells are transplanted into the brain, where they are expected to mature and become a new source of dopamine. For epilepsy in which seizures originate deep in the brain’s temporal lobe and medication does not help, Neurona Therapeutics is testing transplants of neurons that suppress excessive excitation in nearby cells. These neurons are transplanted at a stage when they have stopped dividing but can still migrate from the injection site and integrate into the local neural network.

Early transplants used human fetal tissue, which is difficult to obtain in sufficient quantities and varies between batches in its cellular composition. Stem cells can be expanded, and new batches can be used to grow the required neurons. In a 2011 study, researchers grew dopamine-producing neurons by reproducing the chemical signals that guide midbrain development. These cells survived transplantation into animals with experimentally induced Parkinson’s symptoms. Researchers also had to determine the appropriate stage of maturity: cells transplanted too early may overgrow, while more mature cells survive transplantation less well. Before transplantation, each batch is checked for its cellular composition, including any remaining stem cells capable of forming a tumor.

In a 2025 presentation, cell therapy developer Aspen Neuroscience described a model that uses gene activity in a batch of cells to predict how well they will survive transplantation into rats. The company reports that its other algorithms use images of cells in culture to select areas for robotic processing. Testing of the cell culture system lasted more than 22 months, with operators spending over 6000 hours evaluating it.

Transplant outcomes also depend on delivery: in the TransEuro trial, cells from fetal tissue were transplanted outside the intended target area on both sides of one patient’s brain. A model trained to predict transplant survival therefore needs to account for where and how the cells were delivered.

Mineault proposes using AI to identify changes to cells that would help them survive transplantation into an aged brain, and asks:

Where would you get the data to do this in the first place?

He considers experiments in aged animals and brain tissue maintained outside the body as sources of data on neuronal survival and maturation. Transplant outcomes would be compared with genetic changes introduced into the transplanted cells. A model could then learn from these data to select which changes to test next.

In the Parkinson’s neuron replacement programs Mineault reviews, outcomes are often assessed 18–24 months after surgery because the cells need time to mature and become functionally integrated into the tissue. For clinical testing, he proposes trials that would compare several therapy variants with a shared control group. Testing would proceed in parallel, allowing researchers to evaluate more variants within a single cell maturation period and select those to develop further.