Groundbreaking NTAC Algorithm Proves Synaptic Wiring Alone Can Identify Neuronal Cell Types

JAIST researchers develop NTAC, an automated tool that identifies neuronal cell types using synaptic connectivity alone, outperforming morphology-based methods.

By: AXL Media

Published: Mar 19, 2026, 8:16 AM EDT

Source: Information for this report was sourced from Japan Advanced Institute of Science and Technology

The Shift from Anatomy to Connectivity Fingerprints

Traditionally, neuroscientists have relied on manual morphological classification—the study of a neuron's physical shape and structure—to determine its type. However, as brain mapping transitions into the era of the "connectome," researchers have found that morphology can be deceptive, especially in circuits where different cell types share nearly identical shapes. The development of the Neuronal Type Assignment from Connectivity (NTAC) algorithm marks a paradigm shift by proving that the specific "connectivity fingerprint" of a neuron—how and where it forms synapses—carries sufficient information to identify its biological type without any anatomical data.

Surpassing Morphology-Based Methods in Precision

In comparative tests using fruit fly brain datasets, NTAC significantly outperformed NBLAST, the current industry standard for morphology-based classification. In the optic lobe—a region where neurons are notoriously difficult to distinguish by shape—NTAC achieved over 90% accuracy in identifying cell types. In contrast, shape-based methods struggled to reach 50% accuracy even with significantly more labeled data. This suggests that the functional logic of a brain circuit is embedded more deeply in its wiring instructions than in its physical appearance.

Operational Versatility: Semi-Supervised and Unsupervised Modes

The international research team, led by Dr. Gregory Schwartzman, designed NTAC with two distinct operational modes to handle varying data quality. In its semi-supervised mode, the algorithm requires only a small fraction of pre-labeled neurons to accurately infer the types of thousands of others. For entirely new datasets, the unsupervised mode can group neurons into types based purely on synaptic similarities. Even in the complex full-brain connectome of a fruit fly, which contains thousands of unique cell types, the unsupervised version achieved an encouraging 52% accuracy, far surpassing existing clustering techniques.