A Grain of Brain
The Grain of Brain project is the most detailed wiring-and-activity map of a mammalian brain ever assembled! Incredibly, it is based on analysis of just a square millimetre of mouse visual cortex. This astounding feat was achieved by the Machine Intelligence from Cortical Networks (MICrONS) Consortium. Their work appears in a landmark study that provides an overview of this immensely significant research. Notwithstanding that the brain region they examined is as tiny as a grain of sand.
Using calcium imaging of the visual cortex of living mice to study active neurons, the researchers viewed ~75,000 neurons. Thereafter, they examined ultrathin slices of the same tiny region of the brain with electron microscopy (EM). Lastly, they reconstructed 523 million synapses. It’s a spectacular achievement!
The Allen Institute is at the forefront of the biosciences and one of the major participants in the MICrONS project. In a news release posted on SciTechDaily, they said that this work:
… rivals the Human Genome Project in ambition and scope, using cutting-edge AI, microscopy, and teamwork to map over 200,000 brain cells and millions of synapses. Among many revelations, researchers uncovered surprising new rules for how inhibitory neurons selectively influence others, providing insight into how thought, memory, and disorders like Alzheimer’s might emerge from cellular interactions. This achievement opens the door to a new era in brain science and medical breakthroughs.
A collection of ten open-access papers detailing these findings appears in the esteemed journal Nature, or satellite Nature journals. Have a look at this great introductory video from the Allen Institute, and then read on.
How They Did It: From Light Flashes to Electron Beams
Calcium Imaging
How did they look inside this tiny region of brain? They used what we call calcium imaging. It’s a technique that allows us to see the concentration of calcium inside cells. It uses fluorescent dyes or genetically encoded proteins that change their fluorescence properties in response to calcium binding. The activated cell literally lights up. In other words, this tool allows us to track neuronal activity.
How & Where
The researchers showed mice natural and synthetic videos. They simultaneously recorded responses of ~75 k excitatory neurons across cortical layers 2–5 in primary visual cortex (VISp). They also examined three higher visual areas (VISlm, VISrl, VISal) that integrate visual inputs and other information. Neural traces and behavioural data (pupils, running) capture how each cell reacts to complex stimuli.
Electron Microscopy (EM) and the Grain of Brain
Next, aldehyde fixatives are perfused into the heart of the unfortunate mouse. It dies, and its biochemistry is frozen in time. Thereafter, the fixed mouse brain undergoes dissection.
The same 1.3 × 0.87 × 0.82 mm³ volume of brain that underwent calcium imaging is removed. It’s stained, dehydrated and embedded in epoxy resin. The tiny brain tissue specimen becomes a hard block that can be cut without distortion. It is sliced into ~27,972 ultra-thin (40 nm) sections are placed on an automated tape-collecting microtome. Those sections are not from living tissue—they’re from this stabilised, post-mortem block. Over six months, each of the sections is picked up on grid tape and imaged by automated transmission electron microscopes to build the full 3D volume. They took six months to investigate those specimens, generating 2 petabytes of data.
Proofreading
Convolutional neural networks scan the EM volume and produce an initial segmentation. This automatically groups voxels into neurons, dendrites, axons and synapses. This is analogous to a first‑pass grammar or spell‑checker that tags words and sentences. However, AI-driven automated segmentation and synapse detection are not completely accurate. Consequently, the labelling of thousands of tiny fragments of axon, dendrite, and spine may be incorrect.
During proofreading, expert human proofreaders (and in some cases algorithms proposing edits) merge missing segments back onto their parent neuron and split apart any erroneously fused cells or neurites. They also extend axonal and dendritic branches out to their natural ends, based on the visible EM imagery, until each arbour is as complete as possible.
Without proofreading, the raw, automated reconstructions would contain too many errors—broken branches, false fusions, truncated arbours—to reliably link structure to the functional calcium signals. Proofreading transforms that rough draft into the high‑fidelity connectome at the heart of the MICrONS “grain of brain” experiment.
Expert proofreaders review automated drafts in Neuroglancer‑based tools and perform merge, split and extension edits—just like correcting a sentence structure, fixing typos, and expanding shorthand—to turn the rough AI output into a polished, anatomically correct “connectome.” Incredibly, proofreading amassed 1,046,656 individual edit operations by September 2024.
Grain of Brain to Microscope Co-Registration
Furthermore, and remarkably, expert anchored transforms matched each functionally recorded neuron to its EM counterpart with ~3.8 µm average error, enabling “ground truth” pairing of activity and connectivity for tens of thousands of cells.
Open Access Grain of Brain Tools
Moreover, all data and analysis tools are freely available via MICrONS Explorer, Neuroglancer, DataJoint and cloud volume APIs for anyone to browse, query or download.
https://www.facebook.com/share/v/1FF2ULdqAy
Key Discoveries & Companion Studies
Although still unfolding, the first wave of papers from the April 2025 Nature collection has already revealed:
Cell Type Atlas at Synaptic Scale
Using morphology and connectivity, researchers defined excitatory subtypes across layers and areas, and linked inhibitory cell classes to transcriptomic profiles—uncovering new interneuron categories invisible under light microscopy.
Synaptic Wiring Diagram of a Cortical Column
A complete map of all inputs and outputs for a local column showed “like to like” excitatory connectivity patterns that generalise across layers and areas, validating longstanding theoretical models of cortical processing.
Inhibitory Connectivity Tied to Gene Expression
By matching EM reconstructions with single cell RNA data, the team charted how different inhibitory types (basket, chandelier, Martinotti cells) selectively target excitatory partners—a major step toward for molecularly informed circuit models.
Computational Principles of Visual Integration
AI models trained on this dataset revealed how neurons pool information across space, discovered novel invariances in stimulus encoding, and pointed to general rules governing excitatory linkages both within and between visual areas.
Why It Matters
Grain of Brain: Scale & Resolution
No previous project has bridged function and structure at this scale in a mammalian cortex—until now. Amazingly, the dataset’s cubic millimetre size captures both local microcircuits and long-range cross-area wiring in one shot.
Community Brain Resource
Best of all, by releasing every image, segmentation, trace and synapse map, the MICrONS project invites neuroscientists worldwide to ask new questions—from subcellular organelle distributions to network dynamics in health and disease.
Blueprint for Future Brain Connectomics
The pipeline—combining two photon imaging, petascale EM, machine learning reconstruction and massive proofreading—sets the standard for mapping larger volumes, different brain regions or even other species.
The Grain of Brain Wow Factor
Whether you’re modelling visual computations, probing cell type diversity, or simply marvelling at how half a billion synapses fit in a speck of cortex, this “grain of brain” experiment is your front row ticket to the next era of systems neuroscience. Download the data, explore with Neuroglancer, and see what the brain’s wiring can teach us next.
