South SFV Today

Scientists have made a significant advancement in neuroscience by creating a "digital twin" of the mouse brain. This innovative study, conducted by researchers at Stanford Medicine along with other collaborators, utilizes an artificial intelligence model to emulate the visual processing region of the mouse's brain. This digital twin is trained on extensive datasets of brain activity collected from live mice while they watched movie clips.

The digital twin can predict the responses of thousands of neurons to new visual stimuli. "If you build a model of the brain and it’s very accurate, that means you can do a lot more experiments," stated Andreas Tolias, PhD, a professor at Stanford Medicine and senior author of the study published on April 9 in Nature.

What sets this model apart from previous iterations is its ability to predict neural responses to a wide spectrum of new visual inputs beyond the data it was initially trained on. The model even has the capability to infer anatomical features of each neuron. Lead author Eric Wang, PhD, associated with Baylor College of Medicine, notes the model’s versatility as an example of a foundation model. This new class of AI models learns from large datasets and applies that learning to novel tasks and data types, a process scientists describe as "generalizing outside the training distribution."

The training process involved recording brain activity from mice while they watched action-packed movies, which activated their visual systems. Researchers observed that "mice like movement, which strongly activates their visual system," as Tolias noted. The films chosen aimed to mimic natural settings that mice might observe in their environment.

The amassed data enabled the construction of a core digital twin model, adaptable to simulate individual mice's neural activity from various visual stimuli, including static images. The project's accuracy is attributed to the vast quantity of training data. "They were impressively accurate because they were trained on such large datasets," Tolias remarked.

These digital twins not only replicate neural responses but can also predict the anatomical structure and connection types of neurons within the visual cortex. The accuracy of these predictions was confirmed through electron microscope imaging as part of the broader MICrONS project, which also published its findings in Nature.

The implications of digital twins are far-reaching, potentially revolutionizing research methodologies by offering an inexhaustible resource for experiments. "Experiments that would take years could be completed in hours," Tolias stated, emphasizing the potential acceleration in understanding how brains process information.

Future plans involve extending the modeling to other brain areas and even to animals with more complex cognitive functions, such as primates. "Eventually, I believe it will be possible to build digital twins of at least parts of the human brain," Tolias said, describing this study as merely "the tip of the iceberg."

Contributors to this research include the University of Göttingen and the Allen Institute for Brain Science. The study received funding from multiple organizations, including the Intelligence Advanced Research Projects Activity, the European Research Council, and the National Institute of Mental Health, among others.