Most discriminative stimuli for functional cell type clustering.

Area of Science:

• Neuroscience
• Computational Neuroscience
• Machine Learning

Background:

• Identifying functional cell types in the retina and visual cortex is essential for understanding perception and cognition.
• Current methods for cell-type identification rely on expert knowledge and are biased towards known types, hindering unbiased discovery.
• Unbiased identification of functional neuronal populations remains a challenge in neuroscience.

Approach:

• An optimization-based clustering approach using deep predictive models is proposed to identify functional neuron clusters.
• The method employs an expectation-maximization-like algorithm, alternating between stimulus optimization and cluster reassignment.
• This approach generates Most Discriminative Stimuli (MDS) for unbiased cell-type identification.

Key Points:

• The algorithm successfully recovered functional clusters in mouse retina, marmoset retina, and macaque visual area V4.
• The developed approach demonstrates cross-species and cross-stage applicability in the visual system.
• Most Discriminative Stimuli (MDS) allow for rapid, on-the-fly functional cell-type assignment without complex model training.

Conclusions:

• This novel approach provides a powerful tool for unbiased functional cell-type identification in the brain.
• The interpretable nature of MDS facilitates understanding the stimulus patterns that define specific neuron types.
• The method significantly accelerates experimental workflows, enabling previously time-limited experiments.