How RNA Barcodes Changed What’s Possible in Neuroscience

In April 2026, a technique called Connectome-seq was published in Nature Methods: RNA barcodes assigned to individual neurons, revealing thousands of synaptic connections with unprecedented precision. Combined with the MICrONS mouse cortex map and the fruit fly whole-brain wiring diagram, AI-assisted connectomics has crossed from proof-of-concept into operational science. What can we now see that we couldn’t before — and why does it matter?

Understanding the brain has always come down to a wiring problem. Not the brain’s chemistry, which we have tools for. Not its electrical activity, which we can record. The harder problem is structural: which neuron connects to which, through which pathway, via which type of synapse, at what strength? Without that wiring diagram, neuroscience is reading the output of a circuit without knowing the circuit. Every treatment for every brain disease is being designed without a blueprint.

That is why three developments in the past two years — culminating in the April 2026 publication of Connectome-seq in Nature Methods — represent something genuinely significant. Not because they have solved the wiring problem. The human brain has an estimated 86 billion neurons and 100 trillion synaptic connections. Nobody has solved that. But because the tools for reading brain wiring have crossed a threshold: they have become fast enough, scalable enough, and precise enough to be science rather than an engineering spectacle. The era of functional connectomics — mapping not just structure but the relationship between structure and function — has begun.

The idea of mapping the brain’s wiring diagram is older than modern neuroscience. In 1979, Francis Crick — co-discoverer of DNA’s structure — wrote in Scientific American that the field needed to focus on attainable goals, and that a complete wiring diagram of a model organism’s nervous system was one of them. It took seven more years, but in 1986 Sydney Brenner and colleagues published the first connectome: a complete map of the 302 neurons and approximately 7,000 synaptic connections of the roundworm Caenorhabditis elegans. The map required years of manual analysis of electron microscope images. It remains, four decades later, one of the most cited papers in neuroscience.

The jump from 302 neurons to 130,000 — from C. elegans to the fruit fly — took nearly four decades. It required a fundamental shift in method: from painstaking manual tracing of electron microscope images to AI-assisted automated segmentation, where machine learning algorithms parse the images and trace the fine threads of neuronal processes through three-dimensional volumes of brain tissue. The FlyWire project, completed in 2024 by the Princeton Neuroscience Institute and collaborators, mapped 127,978 neurons and 53 million synapses of an entire adult female Drosophila brain. Nature Methods named electron microscopy-based connectomics its Method of the Year 2025.

“We think that every neuroscience experiment should in some ways be referencing a connectome.”

Thomas Macrina, Princeton Neuroscience Institute — on the MICrONS consortium results, 2025

If the FlyWire fruit fly connectome was the field’s proof of scale, the MICrONS consortium’s 2025 publication was its proof of meaning. The Machine Intelligence from Cortical Networks project, led by teams at the Allen Institute, Baylor College of Medicine, and Princeton University, did something no connectome project had done before: it mapped the wiring and simultaneously recorded the function of the same neurons.

Working from a cubic millimetre of mouse visual cortex — roughly the volume of a grain of sand — the team reconstructed more than 200,000 cells, mapped over 500 million synaptic connections, and traced 4 kilometres of axons through the tissue. Ten studies in the Nature family of journals, published simultaneously in April 2025, described what they found. Researchers built the structural map first. Then, researchers layered calcium imaging data—recordings of neural activity from the very mouse whose brain they had imaged—on top. For the first time, neuroscientists could ask, for individual cells in a mapped circuit: what does this neuron do, what is it connected to, and does that connection pattern explain the function?

The answer, in multiple circuits examined, was yes — with nuance. Early theorists supposed a simple relationship between structure and function, but this is not the case. Some connection patterns predicted functional properties well. Others revealed surprises: cells that appeared, from their anatomy, to be doing one thing turned out to be doing another. The surprises are, in a sense, the point. A wiring diagram does not explain the brain. It makes the questions precise enough to answer.

Both FlyWire and MICrONS use serial-section electron microscopy — a process that involves slicing brain tissue into thousands of ultrathin sections, imaging each section with an electron microscope, and then using AI to reconstruct the three-dimensional structure from the resulting image stack. The results are extraordinary. But the process has a fundamental constraint: it is physical. Tissue must be physically prepared, physically sliced, and physically imaged. It does not scale easily to different experimental conditions, disease states, or genetic backgrounds. Comparing a healthy brain to an Alzheimer’s brain, or a young brain to an aged one, requires rebuilding the entire workflow from scratch.

The paper, published in Nature Methods in April 2026 by Chen, Isakova, Wan, and colleagues, validated Connectome-seq in the mouse pontocerebellar circuit — the pathway connecting the brainstem to the cerebellum, which plays a central role in movement coordination. The results confirmed established connections in the circuit and identified previously uncharacterised synaptic links that had not been detected by prior methods. The molecular markers enriched in connected neurons suggested, for the first time, potential determinants of circuit-specific connectivity: clues to the molecular logic by which neurons choose which other neurons to connect with.

The advance over electron microscopy is not in resolution; EM remains the gold standard for nanoscale structural detail. It is in scalability and information content. Connectome-seq does not require physical slicing. It can be applied to different conditions, different disease states, and different genetic backgrounds in parallel. And because it captures gene expression alongside connectivity, it opens a window that EM cannot: the relationship between what a neuron is (its molecular identity) and who it connects to (its circuit position).

None of the milestones described above would have been achievable without machine learning, and the role of AI in connectomics is worth making explicit. The data volumes involved in electron microscopy connectomics are genuinely extreme. A single cubic millimetre of brain tissue imaged at nanometre resolution produces petabytes of image data — more than most research computing infrastructures could process a decade ago. Manually tracing even a fraction of that volume is not a bottleneck. It is impossible.

AI algorithms — specifically convolutional neural networks trained to segment neuron boundaries, trace axons and dendrites, and identify synaptic structures — turned an impossibility into an engineering challenge. The FlyWire project required the development of CAVE (Connector Annotation Versioning Engine), a system that allowed thousands of neuroscientists worldwide to proofread and refine the AI’s automated segmentation. The MICrONS consortium developed a parallel infrastructure. What resulted in both cases was a human-AI collaboration at a scale without precedent in experimental biology: distributed proofreading by the scientific community, guided by AI, producing structures no human team could have generated alone.

The scientific value of connectomics is increasingly clear. Its medical value is the more urgent argument. The neurological and psychiatric conditions that impose the greatest human burden — Alzheimer’s disease, Parkinson’s disease, autism spectrum disorder, schizophrenia, major depression — are all, at their root, diseases of brain wiring. Neurons connect abnormally, disconnect prematurely, or fail to form the circuits that underlie specific cognitive or behavioural functions. Treatments that address symptoms without understanding circuit-level pathology are, by definition, incomplete.

The MICrONS consortium was explicit about this. They specifically chose the cubic millimetre of mouse visual cortex they mapped because the visual cortex frequently suffers disruptions in neurological conditions like Alzheimer’s disease, autism, and addiction. Researchers at the Allen Institute described the project as designed to “shed light on both form and function within a region that plays a critical role in brain health.” Researchers mapped the wiring diagram of the healthy mouse visual cortex as the baseline against which pathological changes can be measured.

Connectome-seq extends this clinical potential significantly. Because it can be applied across disease states — comparing, say, an Alzheimer’s disease model brain with a healthy one at the circuit level, with gene expression data attached — it creates a path toward understanding not just which circuits are disrupted in neurodegeneration, but which molecular changes drove the disruption. That combination of structural and molecular information, at single-synapse resolution, is what drug discovery researchers need and have not previously had.

A complete human connectome — the full wiring diagram of a human brain at synaptic resolution — remains far beyond current technical capabilities. The fruit fly brain has 130,000 neurons and took a global consortium years to map; the human brain has 86 billion. The difference is not incremental. It is, by current methods, several orders of magnitude beyond reach.

But the history of connectomics since 1986 suggests that the trajectory of progress is not linear. The jump from C. elegans to the fruit fly required new methods, new computational infrastructure, and new institutional models for collaborative science. Connectome-seq represents a new method that sidesteps some of the physical constraints of electron microscopy. Small language models running offline, AI architectures built on connectome data, and parallel advances in nanoscale imaging suggest that the tools being built now are not the tools that will eventually reach human scale.

In the interim, the partial maps matter enormously. A wiring diagram of the circuits involved in a specific disease process — the hippocampal circuits disrupted in Alzheimer’s, the dopaminergic circuits affected in Parkinson’s — does not require a complete human brain map. It requires the right circuit mapped at the right resolution in the right model. The tools to do that exist. What Connectome-seq provides is the ability to do it scalably, comparatively, and with a molecular context attached. That is a significant advance, even at a distance from the full connectome dream.