Google and Janelia Publish a 166,700-Neuron Fruit Fly Connectome Anyone Can Download
On September 3, Cell published "Sexual dimorphism in the complete connectome of the Drosophila male central nervous system." The paper maps the brain, both optic lobes and the ventral nerve cord of one adult male fruit fly, neuron by neuron. The ventral nerve cord is the insect counterpart of the spinal cord. The published abstract counts 166,700 neurons, all proofread and annotated, and sorts them into 11,710 types. Google Research puts the number of synaptic connections at about 125 million and describes the result as the largest brain map to date by neuron count. Janelia says the data and the tools for exploring it are free to researchers worldwide.
The map is called MaleCNS. Janelia's FlyEM team led the project together with the MRC Laboratory of Molecular Biology, the University of Cambridge and Google Research. Janelia imaged the tissue with focused ion beam scanning electron microscopy. Google Research segmented the images, meaning its software traced each neuron through the three-dimensional volume. Teams of experts at Janelia and Cambridge then proofread the tracing and labeled the neurons, connections and cell types. Michał Januszewski and Viren Jain wrote Google's announcement, and both appear on the paper's author list.
A project that drew skepticism
Gerry Rubin, Janelia's founding executive director, assembled the team in 2008. The only finished connectome at that point belonged to the roundworm C. elegans, whose 302 neurons had taken more than a decade of manual work. Rubin recalls critics who predicted that a fly wiring diagram would turn out to be an uninterpretable tangle. The computing side looked just as hard. Early in the project Rubin estimated that proofreading the output of computer vision programs would take 2,000 person-years, and Janelia's own account says standard methods would have required 500 specialists working for a decade.
Jain came to Janelia as a computer vision scientist. His Google Scholar page lists papers on convolutional networks for image denoising in 2008 and for image segmentation in 2010, and his CV places him at Janelia from 2011 to 2014 and at Google from 2014. Google's account says he left because deep learning was making computer vision far more accurate, and that he joined the newly formed Connectomics team in Google Research. The fly was that team's first project. Rubin describes the arrangement as a loop in which Janelia supplied data and proofreading, Google returned automated neuron shapes, and the two sides kept iterating. "Viren built the go-to group in the world for people in the field," Rubin says.
The technical centerpiece was the flood-filling network, which starts at a single pixel and uses a convolutional neural network to find every other pixel belonging to the same neuron. In 2019 the team released a fully automated reconstruction of an entire female fly brain, built from more than forty trillion pixels of electron microscope imagery and thousands of Cloud TPUs. In 2020 came the hemibrain, a human-verified map of half a fly brain with 25,000 neurons and 21 million connections. Google says its methods cut the cost and effort of proofreading by a factor of 20. Janelia says the combined advances in imaging and AI reduced the time and resources needed for a whole-brain connectome by more than 1,000-fold. Both figures come from the institutions that built the tools.
What the comparison with females shows
Until this release every fruit fly brain connectome came from a female, so the male map allows the first synapse-level comparison between the sexes. The abstract reports 8,069 neuron types that are the same in both sexes, 138 that exist in both but wire differently, 289 found only in the male and 71 found only in the female. Sex-specific and dimorphic neurons concentrate in higher brain centers, while the sensory and motor periphery is largely alike in both sexes. Within the higher centers, male-specific connections cluster into hotspots built around male-specific neurons or branches.
The numbers moved between versions. The earlier abstract still listed on Google's publication page gives 166,691 neurons, 11,691 types, and 7,205, 114, 262 and 69 for the four categories above. The published Cell version is the one to cite.
Rubin says this is the first time both sexes of an animal with complex social behavior can be compared, which lets researchers home in on the neurons behind behavioral differences. Because the dataset includes the nerve cord, pathways can be followed from sensory neurons to motor neurons within one dataset. Greg Jefferis of the LMB and Cambridge, who helped lead the project, summed it up this way: "It basically lets us get from eyes to legs in one go." Three companion papers released the same day apply the map to vision, taste and social behavior.
The female side is filling in as well. The 2024 FlyWire brain in Nature has 139,255 neurons, and Google's post refers to a recently published complete female brain and nerve cord map, also in Nature. That post gives no neuron count for the newer female map, so the claim of the largest brain map rests on Google's own statement. With two complete connectomes, researchers can also begin to measure variation between individual animals in regions that are alike in both sexes.
A fly plays Minecraft
The day after release, Georgia Tech graduate student Evan Sinclair Smith posted a demonstration in which the connectome drove a character in Minecraft. According to 404 Media, as relayed by AI Weekly, the post drew millions of views within a day. Versions followed for Doom, Beat Saber, parallel parking and bitcoin trading. Entrepreneur Nico Christie rewired the mating circuit in a variant he called "bi fly," and told 404 Media he wanted the science behind such projects to be more rigorous.
A summary of the demos on daily.dev describes them as approximations that add guessed signal strengths and trained translators to the wiring. A connectome records which neurons contact which, and AI Weekly's report notes that it carries no biochemistry. Google's own post makes a related point about its fish work, describing the connectome as a static resource that must be combined with other information to study plasticity and learning. The map covers one male fly, and Google says two complete connectomes only begin to show individual variability.
Where the methods go next
Google's post names larval zebrafish, in a collaboration with Harvard, as the next full brain. That dataset would combine neural structure and molecular type across an entire vertebrate brain. A separate Fish Fire&Wire dataset pairs whole-brain electrical activity with structure in the same fish. In a Nature study led by Columbia University, the team helped map part of the elephantnose fish hindbrain. Janelia is working on larval zebrafish and the adult fish Danionella cerebrum, and Google is part of a mouse brain effort under the NIH BRAIN CONNECTS program.
Google says adding synthetic neurons to the training data improved the speed and accuracy of its PATHFINDER reconstruction system. That matters because verifying and annotating a fly brain still takes years of human effort. Google also says mapping the 86 billion neurons of a human brain is not yet possible. The largest human sample the team has reconstructed contains 50,000 cells and 150 million synapses, about 1.4 petabytes of data.
Researchers and developers can start at the MaleCNS page hosted by Janelia, which links Neuroglancer views and the Clio annotation interface. Third-party guides describe querying the dataset, labeled male-cns:v1.0, through the neuPrint server, which lists the inputs and outputs of any neuron type with synapse counts. Anyone quoting sizes should use the published Cell figures and the paper's DOI, 10.1016/j.cell.2026.08.015, since the preprint counts differ.