Improving brain models with ZAPBench

Whole-brain activity in a small vertebrate

Traditionally, neuroscientists study neural activity by breaking complex behaviors into smaller parts. To study hunting, for example, they might look at the hunger-sensing capabilities of the cells and organs, the olfactory system that allows an animal to smell their prey, the visual system for tracking, and so on. But complex behaviors almost always involve multiple areas simultaneously, including sensation, decision-making, memory, and movement. To make it even more complicated, neural processing is distributed throughout the brain.

ZAPBench takes a unique approach that focuses on the activity in a vertebrate’s entire brain. Leveraging pioneering work on whole-brain activity recording from our collaborators at Janelia, we built our dataset and benchmark with images captured from the entire brain of the larval zebrafish. We chose the larval zebrafish for several reasons. At only six days old, it is capable of complex tasks that involve motor learning and memory, such as adjusting to moving currents and light conditions, stalking and hunting small prey, and remembering dangerous environments. Furthermore, and most importantly, it is small and transparent, and its entire brain can be imaged under a specialized microscope.

To gather data for our benchmark, our collaborators Alex Chen and Misha Ahrens at HHMI Janelia recorded the fish’s brain activity under a specialized light sheet microscope that uses a laser beam to scan the brain one thin slice at a time and generates a 3D image. The fish was engineered to express GCaMP, a genetically encoded calcium indicator that flashes bright green when it binds to calcium ions that enter active neurons. In order to get a clear image of these proteins as they lit up, the fish was immobilized in a jelly-like substance. To measure its brain’s response to different stimuli, computer generated images were projected around the fish, while the scanning microscope recorded brain activity. A total of two hours of brain activity was recorded in 3D.