Automatic classification and neurotransmitter prediction of synapses in electron microscopy

Area of Science:

• Neuroscience
• Computational Biology
• Electron Microscopy

Background:

• Mapping neural connections (connectomics) using electron microscopy (EM) is crucial but labor-intensive.
• Synapse classification and neurotransmitter prediction are essential for understanding neural circuits.
• Current methods lack automation and detailed functional inference from synapse structure.

Purpose of the Study:

• To develop a deep learning workflow for automated synapse detection and neurotransmitter type prediction in *Ciona intestinalis* EM images.
• To leverage structural information from EM images for accurate synapse classification.
• To enable prediction of neurotransmitter types for *Ciona* neurons, advancing connectome analysis.

Main Methods:

• A deep learning-based workflow utilizing convolutional neural networks (CNNs) was developed.
• Class Activation Maps (CAMs) were employed to visualize and understand CNN decision-making processes.
• The model was trained and validated on *Ciona intestinalis* electron microscopic image data.

Main Results:

• The workflow successfully detects synapses and predicts their neurotransmitter types from EM images.
• Structural features within EM images were found to be predictive of neurotransmitter identity.
• The study successfully differentiated synapses based on their neurotransmitter type, enabling new functional predictions for *Ciona* neurons.

Conclusions:

• Deep learning offers an effective automated solution for synapse detection and classification in EM connectomics.
• Structural analysis of synapses via deep learning provides insights into neuron function and neurotransmitter identity.
• This work significantly advances the understanding of neural circuits in *Ciona intestinalis* by enabling neurotransmitter prediction.