JAIST and Princeton Researchers Build NTAC, an Algorithm That Types Neurons by Their Connections

Researchers from JAIST, Princeton and other centers have presented NTAC, a system that determines a neuron's type by its synaptic connections rather than its outward shape. In experiments on fruit fly connectomes, the algorithm reached accuracy above 90% and completed the task in minutes on an ordinary laptop.

Why shape falls short

Neuron typing has long been bottlenecked by manual work. Experts usually classify cells by morphology: shape, branching, position in tissue. This does not always work. In some brain regions, especially the fruit fly's optic lobe, different neuron types look almost identical. They are hard to tell apart visually, even though they serve different signal-processing circuits. As a result, labeling becomes a slow and expensive process that scales poorly as connectomes grow.

The authors propose looking not at the cell itself but at its "wiring." The logic is simple: two identical wires in a wall are hard to distinguish by appearance, but easy to tell apart if you trace where they lead. Neurons are similar. NTAC uses the pattern of synaptic connections as its main feature, and shows that it reflects a cell's functional nature better than its anatomical silhouette does. This matters especially in dense brain regions with repeating architecture.

How NTAC works

NTAC works in two modes. In the semi-automatic mode, researchers label a small share of neurons in advance, and the model transfers that knowledge to the remaining cells in the same dataset. In the second mode, no labeling is needed at all: the algorithm groups neurons by the similarity of their connections on its own. This matters for large connectomes, where labeling every cell by hand is no longer feasible. The approach can use both partially complete atlases and entirely raw datasets.

• The semi-automatic mode uses a small number of ready-made labels
• The unsupervised mode clusters neurons without hints
• A run takes minutes on a standard laptop
• The method was tested on several fruit fly connectomes

According to the authors, the model was tested on the FlyWire datasets, the optic lobe connectome and ventral nerve cord data. This is not a toy example on a single lab sample, but a comparison across several modern maps of the Drosophila nervous system. That design strengthens the result: the researchers show not a one-off success but an approach that can carry over between different regions, different tasks and different numbers of cell types in a sample.

Results and limits

NTAC performed best where morphological methods most often stumble. In the optic lobe, the semi-automatic variant exceeded 90% accuracy, while the popular shape-based NBLAST approach held at roughly 50%. In fully unsupervised mode, NTAC reached about 70% accuracy in difficult regions, whereas morphological clustering in some cases stayed below 10%. For automatic annotation, that is a very noticeable gap.

"The wiring diagram itself carries enough signal to quickly identify

neuron types."

Also notable: this is not a resource-hungry model built for a data center. The authors stress that NTAC runs on a standard CPU and does not require a supercomputer. For neuroscience this is a practical shift: if connectomes grow from the fly brain to the mouse and then to the human, automated typing will become mandatory. The algorithm has already been used to label thousands of neurons, and the researchers name mouse brain mapping as the next major milestone.

What this means

NTAC does not solve the mapping of the entire human brain, but it removes one of the slowest steps: manual cell typing. If the approach keeps its accuracy on larger organisms, connectomics gains a working tool that could speed up both basic science and the search for neural circuit disorders in disease.