DoGNet: A deep architecture for synapse detection in multiplexed fluorescence images.

Area of Science:

• Neuroscience
• Computational Biology
• Microscopy Imaging

Background:

• Synaptic transmission relies on complex protein interactions within presynaptic vesicles, ion channels, and receptors.
• High-throughput analysis of synaptic structures requires automated and reliable synapse detection methods.
• Current deep learning models, like convolutional neural networks, demand extensive training data for synapse detection.

Purpose of the Study:

• To develop an automated method for synapse detection in multiplexed imaging data.
• To create a neural network architecture that bridges classical computer vision and modern deep learning approaches.
• To enable efficient and accurate synapse classification and phenotypic description using limited training data.

Main Methods:

• Proposing DoGNet, a neural network integrating Difference of Gaussians (DoG) filters with convolutional architectures.
• Optimizing DoGNet for analyzing highly multiplexed fluorescence microscopy data.
• Training and evaluating DoGNet on primary mouse neuronal cultures and mouse cortex tissue slices.

Main Results:

• DoGNet demonstrates superior performance compared to other convolutional networks when trained with limited examples.
• The DoGNet architecture shows efficient transferability between datasets from different research groups.
• Synapse localization by DoGNet facilitates segmentation of synaptic proteins and analysis of their spatial organization.

Conclusions:

• DoGNet offers an efficient and accurate solution for automated synapse detection in multiplexed imaging.
• The method's low training data requirement and transferability make it broadly applicable to neuroscience research.
• DoGNet enables detailed investigation of synaptic protein organization and relative abundance within individual synapses.