An information-geometric framework for statistical inferences in the neural spike train space

Area of Science:

• Computational Neuroscience
• Statistical Inference
• Data-Driven Analysis

Background:

• Traditional analysis of neural spike trains relies on parametric models for temporal evolution.
• Existing methods may not fully capture the variability and distribution of spike trains.

Purpose of the Study:

• To develop a data-driven approach for statistical inference on neural spike trains.
• To introduce novel, generalized metrics for measuring spike train similarity considering time-warping.
• To define and compute statistical measures like mean and variance directly in the function space of spike trains.

Main Methods:

• Development of a parametrized family of penalized L(p) norms as generalized spike train metrics.
• Definition of mean and variance based on these metrics, particularly for p=2 (Euclidean distance).
• Implementation of an efficient recursive algorithm (Matching-Minimization) for computing the sample mean of spike trains.

Main Results:

• The proposed metrics effectively generalize existing spike train distances and account for time-warping.
• The Matching-Minimization algorithm efficiently computes the sample mean for arbitrary spike trains.
• Simulations and experimental data from motor cortex recordings demonstrate the desirable performance of the new framework.

Conclusions:

• The novel framework provides a robust, data-driven method for statistical inference in computational neuroscience.
• The developed metrics and algorithms offer improved analysis of neural spike train variability and distribution.
• This approach successfully captures complex temporal dynamics in neural data.