For almost a decade, Michał Januszewski has been thinking about what fish are thinking about.
Michał is part of Google Research, which has been working with collaborators at HHMI Janelia and Harvard University to build one of the most ambitious datasets in brain activity research yet: a dataset that tracks both the neural activity and nanoscale structure of an entire brain of single larval zebrafish — which could lead to major breakthroughs for how we understand our own brains.
“For years, our team has been really focusing on what’s called connectomics, which deals with the structural mapping of brains — we take very high-resolution pictures of small fragments of brains and try to identify all cells and all the connections between them,” Michał says. “That gives you a static snapshot of the brain as it is at any given moment in time, but doesn’t tell you what the brain is doing when it’s actually alive and thinking.”
So Michał’s team looked to build a new, multimodal dataset that could predict and show neural activity of an organism as it thinks. They chose to start with the zebrafish, which checked several key boxes: It is a vertebrate animal, with more complex brain functions than, say, an insect, and its brain is small enough that the team could get a dataset of the entire brain, instead of just a tiny portion of it.
And — perhaps most importantly — newly hatched zebrafish are almost entirely transparent, allowing the team to use a specialized laser rig to scan nearly two hours of brain activity for more than 70,000 neurons in a live fish’s brain as it reacts to various patterns and stimuli being projected around it.
In April, Google Research released that data as a first-of-its-kind benchmark, called ZAPBench (Zebrafish Activity Prediction Benchmark), which can help advance neuroscience by enabling the development of more accurate AI models that can predict brain activity.
