Google and partners map 166,000-neuron fruit fly brain in largest connectome yet

Google Research, working with HHMI's Janelia Research Campus and other partners, has published a complete wiring diagram of the adult male fruit fly brain and central nervous system. According to a report on dev.to, the map covers roughly 166,000 neurons linked by about 125 million synapses, making it the largest brain map by neuron count published to date.

The milestone is a neuroscience resource rather than an AI product, but it has drawn extra attention because of a side project: DOOMFLY, an experimental pipeline that uses retained connectome data to play the game Doom. The dev.to article draws a sharp line between the two — one is an official research result, the other an unvalidated demonstration — and that distinction is central to understanding what actually happened.

What a complete connectome is

A connectome is a wiring diagram of a nervous system: a record of its neurons and the synapses connecting them. This project mapped the adult male fruit fly's brain and central nervous system at a level that captures both. The dev.to report points to Google's official connectomics announcement and a paper by Januszewski, Jain and collaborators in the journal Cell as the formal record of the work.

Google describes the effort as part of a wider connectomics program that combines AI-assisted reconstruction with human verification, according to the post, and related projects are extending similar mapping approaches to other species, including fish and mice.

The stated research goals include understanding how neural circuits process sensory information and generate behavior, building a foundation for biological and medical research, improving the methods used to reconstruct increasingly complete nervous systems, and informing the long-term design of neuromorphic hardware that takes inspiration from biological neural systems.

Making the data usable

A map of this size is only valuable if scientists can work with it. The program therefore pairs the connectome with visualization and analysis tools, including Neuroglancer and the wider Janelia and FlyWire ecosystem, so researchers can explore, query and test hypotheses against the wiring data rather than simply download a static file.

The DOOMFLY side experiment

DOOMFLY is a separate project hosted in the nftechie/doomfly repository. Per its materials as summarized on dev.to, the system takes frames from Doom, processes them through a simulated fly-brain model, and translates the resulting activity into controls such as movement, turning and shooting. It uses retained MaleCNS v1.0 connectome data and discusses neuroplasticity ideas alongside ongoing validation checks.

The project describes itself as exploratory and non-validated. It is not a living fly playing a video game, not an official Google demonstration, and not evidence of a general-purpose AI. What it does test is whether connectome-scale data can be connected to a real-time control task at all — which forces practical questions about how such a model receives inputs, produces outputs and changes behavior. Those questions sit at the intersection of neuroscience, simulation, AI research and hardware design, but the result so far is a research probe, not a replacement for mainstream machine learning.

The limits of a wiring diagram

A connectome captures wiring relationships, not everything a brain does. As the dev.to article emphasizes, a full account of brain function also involves dynamics and other biological processes, so the map is a major input to future research rather than a finished explanation of intelligence. There are also no performance, efficiency, pricing or availability figures for any connectome-derived system, so claims about cheaper or faster AI from this work would be premature.

Why it matters

The significance is twofold. First, the infrastructure for reconstructing complete nervous systems has advanced considerably: mapping 166,000 neurons and 125 million synapses, with AI doing the heavy lifting and humans verifying the results, provides a template that can scale toward more complex organisms.

Second, Google explicitly frames connectomics as work that may inform neuromorphic hardware design. Biological neural circuits embody a very different model of computation from today's processors, and a complete wiring diagram gives researchers a real circuit to study rather than an abstraction. For AI and hardware teams, the practical lesson is to separate deployable tools from long-term signals: nothing here ships tomorrow, but the map is likely to shape research into future architectures and specialized chips — and the fruit fly brain is now the most complete reference point available.