DataJoint: Data and Operations Infrastructure for the MICrONS Brain Study

Launched in 2016 by the U.S. Intelligence Advanced Research Projects Activity (IARPA), the MICrONS Program set out to reverse-engineer the brain’s computational strategies by linking what neurons do (function) to how they’re connected (structure).

Three institutions led the effort: Baylor College of Medicine (Tolias Lab), Princeton University (Seung Lab), and the Allen Institute for Brain Science. Together, they formed the core of a multidisciplinary team that spanned more than a dozen institutions — including Harvard, JHU Applied Physics Laboratory, MIT, Carnegie Mellon, and others — uniting top minds in systems neuroscience, imaging, AI, and data science.

MICrONS dataset overview. Nature, 2025
(DOI 10.1038/d41586-025-00908-4)

The ambition of MICrONS was unmatched: to map and understand how thousands of neurons work together — creating a digital twin of brain structure and function at millisecond and nanometer resolution.

This required three major scientific feats:

Functional Imaging — Measuring the activity of tens of thousands of neurons in response to naturalistic video stimuli, over weeks of live behavior.
‍Structural Reconstruction — Imaging the same brain volume using serial section electron microscopy, producing over a petabyte of raw imaging data.
‍Connectome Mapping — Segmenting and reconstructing the entire wiring diagram of every cell and synapse in the imaged region.This effort produced datasets capturing activity, morphology, and connectivity from the same brain tissue — offering both functional and structural data for a complete volume.

This effort produced datasets capturing activity, morphology, and connectivity from the same brain tissue — offering both functional and structural data for a complete volume.

This was not just a moonshot — it was a systems neuroscience milestone. MICrONS redefined what it means to understand the brain, and laid the groundwork for the next generation of AI systems inspired by real neural architectures.

MICrONS by the Numbers

1 mm³ of mouse visual cortex
82,000 neurons
500 million synapses
Petabytes of multi-modal data
9 years from start to publication

From Live Recordings to Unified Discovery

Imaging Neuronal Activity at Scale

This produced one of the largest and most detailed datasets of in vivo brain activity ever recorded — setting the stage for subsequent structural analysis and registration.

DataJoint powered the functional imaging study that captured how neurons in the living brain respond to natural stimuli. Using multiphoton mesoscopy, the team recorded calcium signals from tens of thousands of neurons in transgenic mice as they viewed natural scenes, virtual reality environments, and synthetic visual patterns.

A Pipeline Built for Collaboration

Scientists could query the latest results to monitor progress, share insights, and make decisions without bottlenecks. DataJoint automated routine computations and issued real-time updates, keeping experimental and analysis workflows in sync. While experimentalists refined imaging conditions, data scientists trained and tested models — all on a living, shared dataset.

The functional data pipeline, built in DataJoint, integrated contributions from multiple collaborating scientists in US and EU institutions. It orchestrated acquisition, processing, and analysis of neuronal activity and behavior — enabling rapid iteration and shared insight within the team.

“We found that movie clips stimulated the sensory circuits far more intensely and broadly than classical synthetic stimuli.”

— Dimitri Yatsenko, PhD

Quality Without Compromise

Signal quality was a top priority. The pipeline quantified key indicators of experimental success — imaging clarity, tissue motion, brain state, laser power, and calcium indicator response — for every session. DataJoint continuously tracked these metrics, and the team used them to guide methodological improvements throughout the project.

As protocols evolved, these controls ensured consistency across hundreds of imaging sessions, leading to the highest standard of reliability in large-scale functional recordings.

In vivo recorded data on inputs (visual stimulus, eye position, locomotion, and pupil size) and outputs (neural activity) trains an artificial neural network model to generate in silico responses. See Foundation model of neural activity predicts response to new stimulus types, Figure 1.

Registering Structure and Function

As the two other principal teams completed their analyses, DataJoint pipelines pulled in selected information from their morphology and connectivity analysis — enabling scientists to query functional data in the context of structural features.

Before this, such integrated analysis was largely hypothetical. With DataJoint, it became real.

This unified pipeline enabled scientists to write simple, expressive queries in DataJoint Python that span both structure and function — for example, identifying the shape of a neuron and its input connections, then querying how it responded to specific stimuli.

Understanding Neural Intelligence — Biological or Artificial
Professor Andreas Tolias, Stanford Medical School
DataJoint Co-Founder

Open Access to a Landmark Dataset

A substantial portion of the MICrONS functional dataset is now public — shared as a structured, queryable pipeline with functional data on the “Platinum Mouse.”

But publishing this pipeline was no small task. The original “Cajal” pipeline developed by the BCM team and used internally for years was extraordinarily rich and complex — a living system developed over a decade of experimentation and collaboration.

To make it accessible, the team extracted a simplified version into a clean, queryable structure with processed data, built specifically for public release. This version enables outside researchers to interact with the data meaningfully, even without reproducing every processing step.

The Microns Explorer website documents additional datasets and visualizations from the morphology and connectivity studies, with tutorials and tools for accessing and analyzing these data in their native platforms.

Three Ways to Access the MICrONS Pipeline

Self-service

See instructions on how downloading the simplified functional pipeline from DANDI archive and setting up the necessary systems to access and use the pipeline.

Freely hosted

A live instance of the simplified functional pipeline is hosted by DataJoint and freely available (runs in Github Codespaces). See the GitHub tutorial project for more information.

Full-service

You can build MICrONS data into your own custom pipeline and extend the analysis in DataJoint's full-featured analysis environment. To learn more, please ask our team.

The Full Pipeline: Expanding the Scope

An ongoing collaboration is now using a more complete version of the MICrONS functional dataset — including multiple animals, raw inputs, all intermediate processing steps — to reproduce and extend the original analysis.

Teams led by Dimitri Yatsenko (DataJoint Inc.), Paul G. Fahey (Stanford / Tolias Lab), and Nima Dehghani (MIT / Senseable Intelligence Group), including several international students from Neuromatch, are currently using DataJoint to recreate the entire flow of the MICrONS functional study. Their pipeline preserves every data transformation and includes source data from many mice beyond the “Platinum Mouse” featured in the published dataset. The goal is to enable higher-significance analysis and open the door to new scientific discoveries

If you’re interested in contributing to this effort or in accessing the deeper dataset and reproducible environment, please contact the DataJoint team to learn more.

Not Just for Moonshots

MICrONS did what Crick called “the impossible.”

DataJoint helped make it possible.

In 2020, NIH began funding DataJoint's evolution. Today, we offer a cost-effective platform that brings all the benefits to any lab of the system that powered one of the most ambitious projects in history.

You don’t need a massive budget or a team of engineers.

You need a platform built for scientific impact.

Get Started with DataJoint