Analysis of Neuronal Spike Trains, Deconstructed

Area of Science:

• Neuroscience
• Computational Neuroscience
• Quantitative Biology

Background:

• Neuronal firing encodes sensory information and drives motor behavior.
• Developing predictive models linking neural activity to external variables is a key goal in neuroscience.

Purpose of the Study:

• To review and contrast analytical tools for building predictive models of neuronal firing.
• To focus on models that extract low-dimensional representations from external variables and incorporate neural history.
• To illustrate techniques with complex datasets, including those with strong correlations.

Main Methods:

• Review of analytical tools for relating sensory/motor streams to neuronal firing.
• Focus on models comparing external variables to feature vectors for low-dimensional representation.
• Incorporation of spiking history and nonlinear transformations for spike prediction.
• Application to datasets of varying complexity, including natural stimuli and movement data.
• Use of spectral correlation to compare model performance.

Main Results:

• Different analytical tools are presented and contrasted for their effectiveness.
• Techniques are illustrated using complex datasets, demonstrating their applicability.
• Model fitting in the presence of strong correlations (natural stimuli, movement) is addressed.
• Spectral correlation provides a metric for evaluating the success of different modeling approaches.

Conclusions:

• A range of analytical tools exist for modeling neuronal firing based on external variables.
• Model selection and fitting require careful consideration of data complexity and correlations.
• Spectral correlation is a valuable method for assessing the performance of predictive neural models.