Modeling functional cell types in spike train data.

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Neuroscience

Background:

• Accurate computational models of neuronal activity are crucial for understanding neural circuit function.
• Current single-cell models often lack refinement, limiting their interpretability and predictive power.
• Functional cell typing offers a promising avenue to enhance these models.

Purpose of the Study:

• To develop and validate a novel computational framework for inferring functional cell types and refining single-cell models simultaneously.
• To improve the accuracy and interpretability of models of neural responses by leveraging shared functional properties across neurons.
• To assess the generalizability and biological relevance of discovered cell-type clusters.

Main Methods:

• Developed a hierarchical generative model for cell types, single-cell parameters, and neural responses.
• Derived an expectation-maximization algorithm with variational inference to maximize neural recording likelihood.
• Applied the simultaneous method to simulated data and in vitro neural recordings from mouse primary visual cortex.

Main Results:

• The method accurately recovered ground truth parameters from simulated data.
• Applied to mouse visual cortex data, the approach significantly improved single-cell activity prediction.
• Discovered cell-type clusters were well-separated, generalizable, and amenable to interpretation.
• Cluster memberships showed correlations with locational, morphological, and transcriptomic data.

Conclusions:

• Explicitly modeling shared functional properties through cell typing can substantially improve models of neural responses.
• The developed simultaneous inference method provides an effective tool for cell-type discovery and model refinement in neuroscience.
• This approach holds potential for advancing our understanding of neural computation and circuit organization.