“We have shown that it is possible to automatically classify neuron types across the brain of Drosophila (an essential model organism in neuroscience) with high accuracy, particularly on the visual system, using only their connectivity patterns,” explains researcher David Garcia, a Serra Húnter professor at the Department of Computer Science of the Universitat Politècnica de Catalunya - BarcelonaTech (UPC). “Our method requires far less information: it simply uses the graph of connections between neurons and performs the classification automatically, within minutes and with high precision. Until now, we didn’t know this was possible,” he notes.

To date, this classification was carried out manually by expert neuroscientists through a painstaking process that required additional morphological and anatomical information (the shape and position of neurons). According to the published study, this task can now be automated with high reliability using only synaptic connectivity patterns. This represents a paradigm shift in the analysis of large connectomics datasets.

“The direct application of our work is precisely the automatic classification of neurons in new connectomes, which is a fundamental step in neuroscience to understand how brain circuits work. This means that future connectomes could be analysed automatically within hours instead of requiring long periods of manual work,” explains Garcia. He adds: “In the longer term, this approach could have applications in other domains beyond neuroscience, for example in graph mining, although this still remains to be explored.”

In recent years, projects such as FlyWire, a global effort that has reconstructed the whole brain connectome for the fruit fly using electron microscopy, have produced detailed maps of all neuronal connections. The next step, identifying the cell type to which each neuron belongs, remained a bottleneck that required months of manual annotation, the researcher notes.

The published study introduces a method, NTAC, capable of assigning neuronal types based solely on synaptic connectivity with very high accuracy. This contribution emerges at a time of rapid expansion in connectome datasets and a growing need for automatic, scalable tools.

NTAC is presented in two forms: a semi-supervised one that leverages a small fraction of labelled neurons and an unsupervised one that requires no labels at all. Both deliver high‑precision results and run efficiently even on conventional computers, demonstrating the potential of connectivity as a basis for identifying neuronal types without relying on morphological features.

The conjecture that connectivity alone might be sufficient to determine neuronal identity was explicitly proposed some years ago by scientist Sebastian Seung in his book Connectome.

The new study now confirms this idea with data and methods that are publicly accessible to the scientific community. The paper “Neuronal type assignment from connectivity” has been published open access in Nature Communications and represents a significant advance in the study of the brain and neuronal classification.

It is the result of an international collaboration between UPC researcher David Garcia, who last year received the second prize in the Juan López de Peñalver Award from the Spanish Royal Academy of Engineering, and scientists from the Princeton Neuroscience Institute, the Japan Advanced Institute of Science and Technology and the University of Edinburgh. Nature Communications additionally highlights the study in its Focus section as one of the most noteworthy brain‑related papers recently published..

More information

• NTAC: Neuronal type assignment from connectivity