A frequency-resolved mutual information rate and its application to neural systems

Area of Science:

• Computational Neuroscience
• Information Theory
• Neural Coding

Background:

• Quantifying information flow in sensory neurons is challenging, with Shannon's information theory yielding only a single mutual information rate.
• Existing methods for frequency-resolved information analysis rely on approximations like the coherence function, whose accuracy is often unknown.
• Understanding how neurons selectively encode slow versus fast stimulus components (low-pass vs. high-pass filtering) is crucial but limited by current analytical tools.

Purpose of the Study:

• To develop an assumption-free method for measuring frequency-resolved information rates in neural responses to time-dependent stimuli.
• To apply this new method to paradigmatic neural firing models, including Poisson processes and integrate-and-fire neurons.
• To compare the results with previous coherence-based estimates and assess the limitations of response-response coherence as an upper bound for information rate.

Main Methods:

• Development of a novel, assumption-free analytical technique to calculate frequency-resolved information rates.
• Application of the method to three distinct neural models: inhomogeneous Poisson process, stochastic integrate-and-fire neuron, and two coupled integrator neurons.
• Analysis of information filtering properties (broadband, low-pass, band-pass) across different frequency components of a time-dependent Gaussian stimulus.

Main Results:

• Inhomogeneous Poisson processes and integrate-and-fire neurons act as broadband and low-pass information filters, respectively, consistent with prior coherence estimates.
• The band-pass information filtering observed in synchronous spikes of coupled neurons is confirmed by the new method in certain parameter ranges.
• The study demonstrates cases where response-response coherence can overestimate the actual information rate, highlighting its limitations.

Conclusions:

• The developed assumption-free method provides a more accurate and reliable way to measure frequency-resolved information rates in neural systems.
• This approach offers deeper insights into how neural circuits process and filter time-varying sensory information at different timescales.
• The findings underscore the importance of using precise methods for analyzing neural coding and caution against over-reliance on coherence measures.