PT  - JOURNAL ARTICLE
AU  - Matthew R Whiteway
AU  - Karolina Socha
AU  - Vincent Bonin
AU  - Daniel A Butts
TI  - Characterizing the nonlinear structure of shared variability in cortical neuron populations using neural networks
AID  - 10.1101/407858
DP  - 2018 Jan 01
TA  - bioRxiv
PG  - 407858
4099  - http://biorxiv.org/content/early/2018/09/04/407858.short
4100  - http://biorxiv.org/content/early/2018/09/04/407858.full
AB  - Sensory neurons often have variable responses to repeated presentations of the same stimulus. Simultaneous recordings of neural populations demonstrate that such variability is often shared across many neurons, and thus cannot be simply averaged away. Understanding the effects of this shared variability on neural coding requires an understanding of what the common drivers of variability are, and how they are related to each other and to stimulus processing. Latent variable models offer an approach for characterizing the structure of this shared variability in neural recordings, though most previous modeling approaches have either been linear or had to make very restrictive assumptions about the nonlinear structure. Here we demonstrate the use of an autoencoder neural network as a general means to fit nonlinear latent variable models for analyzing large-scale neural recordings. We first present a general nonlinear latent variable model that is agnostic to the stimulus tuning properties of the individual neurons, and is hence well suited for exploring neural populations whose tuning properties are not well characterized. This motivates a second class of model, the Generalized Affine Model, which simultaneously determines each neuron’s stimulus selectivity and a set of latent variables that modulate these stimulus responses both additively and multiplicatively. Despite the general nonlinear relationships that can be uncovered with our approaches, we find that population variability in anesthetized primary visual cortex (V1) is best described by a single additive and single multiplicative latent variable. However, this is not generally the case, and to contrast the V1 results we apply the same models to recordings in awake macaque prefrontal cortex, where more general nonlinearities are needed to compactly describe the population response variability. These results thus demonstrate how nonlinear latent variable models can be used to describe population variability, and suggest that a range of methods is necessary to study different brain regions under different experimental conditions.