Dethroning the Fano Factor: A Flexible, Model-Based Approach to Partitioning Neural Variability

Area of Science:

• Computational neuroscience
• Systems neuroscience
• Neuronal variability modeling

Background:

• Neurons exhibit significant trial-to-trial variability, often overdispersed compared to Poisson distributions.
• Previous models proposed multiplicative gain, leading to quadratic mean-variance relationships.

Purpose of the Study:

• To propose a more flexible model family for neuronal variability.
• To investigate diverse mean-variance relationships beyond quadratic.
• To improve understanding of neuronal firing rate variability.

Main Methods:

• Developed a model with additive Gaussian noise transformed nonlinearly into a Poisson spike rate.
• Explored various nonlinear transformations to generate different mean-variance relationships.
• Implemented an efficient method for fitting the model to neural data.

Main Results:

• A rectified squaring nonlinearity yielded a linear mean-variance function (constant Fano factor).
• A majority of V1 neurons showed nonquadratic mean-variance relationships, better fitting the new models.
• The model improved Bayesian adaptive stimulus selection for tuning curve estimation.

Conclusions:

• Neuronal variability is more diverse than previously modeled.
• Flexible models capture a wider range of mean-variance relationships in neural data.
• Accounting for overdispersion enhances neurophysiological experiments and data analysis.