Classification and analysis of retinal interneurons by computational structure under natural scenes

Area of Science:

• Neuroscience
• Computational Neuroscience
• Vision Science

Background:

• Retinal amacrine cells are diverse inhibitory neurons crucial for visual processing.
• Functional classification of these cells under natural stimuli remains challenging.

Approach:

• Utilized optical recording and an interpretable convolutional neural network (CNN) model to analyze mouse amacrine cell responses to natural scenes.
• Applied interpretable machine learning to identify model interneurons presynaptic to amacrine cells, analogous to bipolar cell inputs.
• Clustered amacrine cells based on their model presynaptic inputs rather than direct responses to natural scenes.

Key Points:

• Identified approximately seven distinct amacrine cell groups based on their input structure.
• Demonstrated that distributed, rather than dedicated, synaptic inputs shape the natural scene responses of different amacrine cell types.
• Revealed distinct transient and sustained network response modes to simple visual stimuli.

Conclusions:

• This study provides a new method for functionally classifying retinal amacrine cells.
• The findings illuminate the computational principles governing amacrine cell responses to natural scenes.
• Generated testable hypotheses regarding synaptic input organization and function in the retina.