Filter-based models of suppression in retinal ganglion cells: Comparison and generalization across species and stimuli

TL;DR

This paper compares computational models of suppression in retinal ganglion cells across species and stimuli, finding that subtractive and divisive models perform best under different conditions.

Contribution

The study introduces a comparative analysis of suppression models in retinal ganglion cells across species and stimulus types.

Findings

Subtractive and divisive suppression models outperform feedback and linear-nonlinear models in predicting ganglion cell responses.

Divisive suppression models generalize better across temporal frequencies and contrast, while subtractive models excel for slow dynamics.

Suppression inclusion improves model performance compared to models without suppression.

Abstract

The dichotomy of excitation and suppression is one of the canonical mechanisms explaining the complexity of neural activity. Computational models of the interplay of excitation and suppression in single neurons aim at investigating how this interaction affects a neuron’s spiking responses and shapes the encoding of sensory stimuli. Here, we compare the performance of three filter-based stimulus-encoding models for predicting retinal ganglion cell responses recorded from axolotl, mouse, and marmoset retina to different types of temporally varying visual stimuli. Suppression in these models is implemented via subtractive or divisive interactions of stimulus filters or by a response-driven feedback module. For the majority of ganglion cells, the subtractive and divisive models perform similarly and outperform the feedback model as well as a linear-nonlinear (LN) model with no suppression. Comparison between the subtractive and the divisive model depends on cell type, species, and stimulus components, with the divisive model generalizing best across temporal stimulus frequencies and visual contrast and the subtractive model capturing in particular responses for slow temporal stimulus dynamics and for slow axolotl cells. Overall, we conclude that the divisive and subtractive models are well suited for capturing interactions of excitation and suppression in ganglion cells and perform best for different temporal regimes of these interactions.

The spiking activity of neurons throughout the nervous system is shaped by how excitatory signals interact with mechanisms of suppression, such as input from inhibitory neurons or synaptic fatigue. For example, suppressive signals can influence a neuron’s sensitivity to different sensory stimuli, the temporal structure of its activity, or how it adapts its responsiveness under different ranges of input intensity. Conceptually, one often distinguishes whether the suppression acts on the excitatory signal in a subtractive fashion, in a divisive fashion, or as a feedback signal whose strength depends on the previously elicited activity. Here, we investigated three corresponding computational models with similar overall structure but different suppressive interactions. To evaluate the models in the context of visual stimulus encoding, we assessed their ability to predict measured light responses of retinal ganglion cells, the output neurons of the vertebrate retina, in different species (axolotl, mouse, marmoset). We found that including suppression was generally better than not including suppression and that the subtractive and divisive models overall outperformed the feedback model. Moreover, the former two models were best suited for different conditions, with divisive suppression excelling for rapid stimulus kinetics and subtractive suppression better capturing slow response modulations.

Figures