Connectivity Is All You Need: Inferring Neuronal Types with NTAC.

Area of Science:

• Neuroscience
• Computational Biology
• Systems Biology

Background:

• Electron microscopy and computer vision enable large-scale connectome mapping.
• Accurate neuronal cell type identification is crucial for understanding brain function.
• Traditional cell type identification methods are labor-intensive and rely on multiple features.

Purpose of the Study:

• To develop an automated method for neuronal cell type classification based solely on synaptic connectivity.
• To validate the hypothesis that synaptic connectivity is a primary determinant of neuronal cell types.
• To introduce and evaluate NTAC (Neuronal Type Assignment from Connectivity) in both semi-supervised and unsupervised settings.

Main Methods:

• Developed NTAC, a graph-based learning method for semi-supervised neuronal cell type assignment using synaptic data.
• Introduced approximate equitable partitioning and a heuristic for unsupervised NTAC.
• Utilized NTAC's semi-supervised algorithm as a subroutine in the unsupervised approach.
• Evaluated NTAC on multiple fruit fly connectomes (optic lobes, central brain, nerve cord).

Main Results:

• The semi-supervised NTAC achieved over 95% accuracy on the fruit fly visual system with only 2% labeled neurons.
• NTAC's semi-supervised approach outperformed morphology-based methods in accuracy and labeling requirements.
• The unsupervised NTAC achieved approximately 70% accuracy, significantly outperforming morphology-based parallels.
• Results provide strong evidence that synaptic connectivity alone can define neuronal cell types.

Conclusions:

• NTAC offers an efficient and accurate approach to neuronal cell type identification using synaptic connectivity.
• Both semi-supervised and unsupervised NTAC variants demonstrate the power of connectivity-based classification.
• This work advances automated analysis of large-scale connectomic datasets.