Deep Learning-Based Classification of GAD67-Positive Neurons Without the Immunosignal

Area of Science:

• Neuroscience
• Computational Biology
• Cell Biology

Background:

• Neural circuits comprise excitatory neurons and GABAergic interneurons, crucial for information processing.
• Interneurons are less abundant than excitatory neurons in regions like the neocortex and hippocampus.
• Traditional quantification of interneurons relies on markers like GAD67 (glutamic acid decarboxylase), with varying expression of other proteins like NeuN (neuronal marker).

Purpose of the Study:

• To determine if GAD67-immunopositive neurons can be identified using NeuN and Nissl fluorescence signals.
• To develop and validate a deep learning algorithm for unbiased cell-type classification without GAD67 staining.

Main Methods:

• Neurons in mouse S1 and M1 cortex layers 2/3 were stained and manually classified using GAD67 immunosignals.
• A custom deep learning algorithm was trained to detect GAD67-positive neurons using only Nissl and NeuN signals.
• Performance was compared against classic machine learning methods.

Main Results:

• The deep learning model successfully performed binary classification of neurons using Nissl and NeuN signals alone.
• The deep learning approach demonstrated superior performance compared to classic machine learning methods.
• Visualization of the algorithm's hidden layer offered insights into unbiased cell-type classification criteria.

Conclusions:

• A novel deep learning-based platform enables accurate identification of GAD67-positive neurons without direct GAD67 immunostaining.
• This method provides a more objective and efficient approach to neuronal cell-type classification.
• The findings pave the way for unbiased criteria in neuroscience research.