Google maps entire male fruit fly brain with 166,000 neurons

To put it plainly, the Google Research team took AI technology originally built for recognizing objects in photos and repurposed it to stitch millions of brain cross-section images into a 3D reconstruction. The team produced the first fully automated reconstruction of a female fruit fly brain in 2019, then completed a human-verified map of half that brain in 2020. This new complete map of the male fruit fly brain is the result of nearly a decade of continuous work building on that foundation.

The map covers not just the brain but also the ventral nerve cord, the fly's equivalent of a spinal cord. That means it offers a resource for studying how the brain controls body movement, not just cognition. The full dataset is described in an announcement from HHMI Janelia, which notes that a team of Janelia specialists manually verified and annotated every individual neural structure in the map.

Why start with the fruit fly brain

You might wonder why mapping a fruit fly's brain matters. The human brain has 86 billion neurons, far beyond what current technology can map in full. So scientists are starting with smaller organisms like fruit flies, using AI to map their brains first, in hopes of understanding how animals respond to stimuli and how damaged neural pathways might be repaired.

Fruit flies already have a strong track record in genetics research, having contributed to multiple Nobel Prizes. Their standardized behavioral patterns and short life cycles have made them a core model organism for genetics, and this research represents an attempt to extend that role into neuroscience. The underlying premise is that even brains from very different species share a lot of common structure.

AI stitches electron microscope images into 3D

Connectomics work involves slicing a brain into millions of thin cross-sections, photographing each one, and then using computers and AI to stitch those images back together. Google's research team has been building a system to reconstruct 3D neural structures from flat electron microscope images, and one of its core technologies is a flood-filling network built on convolutional neural networks, which starts from a single pixel and identifies every other pixel belonging to the same object.

In 2019, the team released a fully automated reconstruction of the female fruit fly brain. In 2020, they completed a human-verified map of half of that brain. More recently, the team said it mixed synthetic neurons into its training data to boost both the speed and accuracy of its newest reconstruction system, called PATHFINDER.

Comparing against the female map to study sex differences

This new male brain map also complements an earlier connectome of the female fruit fly brain and central nervous system. Having maps of both sexes lets researchers compare regions where the neurons differ, which helps in studying the biological mechanisms behind courtship and aggression behaviors in fruit flies. Conversely, comparing similar regions across the two sexes can reveal where individual variation comes from. In one case, AI-based 3D reconstruction precisely identified a structural difference in which a single male neuron had two extra branches compared to its female counterpart.

Next targets: zebrafish and mouse brains

The research team is now extending connectomics to vertebrates with spinal cords. In a study led by Columbia University and published in Nature this week, researchers mapped part of the signal-processing region in the hindbrain of the elephantnose fish. The team said it's the first case of combining a static connectome with other data to reveal the learning mechanism of a vertebrate brain. Zebrafish, whose larvae have transparent bodies that let researchers observe neural activity during experiments, are one of the few vertebrates whose entire brain can be mapped with current technology. The team's next paper, done jointly with Harvard, is expected to be the first vertebrate forebrain dataset that includes both neural structure and molecular cell types. Work mapping part of the mouse brain is planned to follow after that.

Timing	Subject	Neuron count	Connection count	Verification status
2019	Female fruit fly brain	-	-	Fully automated, unverified
2020	Half of female fruit fly brain	25,000	21 million	Human-verified
2026	Male fruit fly brain + CNS	166,000+	125 million	Human-verified

Where to see the data

The completed male fruit fly connectome can be viewed at male-cns.janelia.org, and the raw imaging data is available on Janelia's FlyEM project page. Visualization runs through Neuroglancer, an open-source tool built by Google that's designed to let researchers browse and download large, multidimensional datasets directly. Three follow-up papers released alongside this one reportedly show how the connectome is already being used in research on the visual system, taste, and social behavior.

Editor's view

This announcement reads more clearly as an infrastructure investment than as a product launch from an AI company. What Google Research is doing here isn't selling a chatbot or an image generator — it's spent nearly a decade refining a reconstruction engine that turns piles of electron microscope images into 3D neural network maps. Trace the lineage from flood-filling networks to PATHFINDER, and this isn't a single paper but a cumulative body of work from the same team steadily pushing precision higher.

The generational comparison makes the progress tangible. In 2019, the reconstruction was automated but unverified. In 2020, verifying a half-brain map of just 25,000 neurons by hand took considerable time. This time, the team delivered a fully verified map covering more than six times as many neurons — 166,000 — spanning not just the brain but the nerve cord as well. The same team, using the same basic methodology, expanded what it could handle by that much in just six years.

In practical terms, the immediate beneficiaries of this dataset are researchers in neuroscience, pharma, and medicine. Teams working on brain science or neurological disease, including in Korea, can open this publicly available dataset directly through Neuroglancer to test their own hypotheses. For the AI field, the more interesting lesson may not be the image reconstruction technology itself but how it was transplanted into an entirely different domain — biology and medicine. It's also worth noting that wiring diagrams like this one serve as foundational material for understanding the biological mechanisms behind conditions like Alzheimer's, depression, and schizophrenia.

In the coming months, a complete zebrafish brain map, being developed jointly with Harvard, could be the next milestone to watch. As this progression continues from fruit flies toward vertebrates, the goal of mapping the human brain's 86 billion neurons remains far off — but the gaps in between are being filled in, one step at a time.