Optimal firing rate estimation.

Area of Science:

• Computational Neuroscience
• Neural Coding
• Information Theory

Background:

• Evaluating predictive models of spiking neuron behavior is crucial for understanding neural computation.
• Existing methods may not fully capture the information content conveyed by individual neural spikes.

Purpose of the Study:

• To define and apply a novel information-theoretic measure, information gain per spike (Is), for assessing the quality of spiking neuron predictive models.
• To optimize parameters of Gaussian smoothing filters used for neural firing rate estimation using the Is criterion.

Main Methods:

• Defined information gain per spike (Is) as a measure comparing model information to average firing rate prediction.
• Applied a maximum Is criterion to optimize Gaussian smoothing filter bandwidth for firing rate estimation.
• Validated the method using data from bullfrog vestibular semicircular canal neurons and simulated integrate-and-fire neurons.

Main Results:

• The Is measure quantifies the added information from a model beyond simple average firing rate predictions.
• Optimal bandwidth for firing rate estimation was found to be similar to the average firing rate across tested neuron types.
• Precise timing and average rate models were identified as suboptimal, performing poorly compared to optimized models.

Conclusions:

• The information gain per spike (Is) provides a robust metric for evaluating spiking neuron models.
• Optimized firing rate estimation using Is enhances the understanding of neural information transmission.
• Bullfrog semicircular canal sensory neurons appear to transmit approximately 1 bit of stimulus-related information per spike.