Fast computation of the statistical significance test for spatio-temporal receptive field estimates obtained using spike-triggered averaging of binary pseudo-random sequences

TL;DR

This paper introduces a faster method for computing the statistical significance of spatio-temporal receptive field estimates obtained via spike-triggered averaging, using a Normal approximation to reduce computation time without sacrificing accuracy.

Contribution

The authors develop an accelerated significance testing method for STRF estimates using a Normal approximation, significantly reducing computation time while maintaining accuracy.

Findings

The approximate method yields the same significance thresholds as the exact method.

Significant reduction in computation time, from minutes to seconds or minutes.

Validation on mouse retinal ganglion cell data confirms effectiveness.

Abstract

Background: Spatio-temporal receptive fields (STRF) of visual neurons are often estimated using spike-triggered averaging of binary pseudo-random stimulus sequences. The stimuli are visual displays that contain black and white pixels that flicker randomly at a fixed frame rate without any spatial or temporal correlation. The spike train of a visual neuron, such as a retinal ganglion cell, is recorded simultaneously with the stimulus presentation. The neuron's STRF is estimated by averaging the stimulus frames that coincide with spikes at fixed latencies. Recently, an exact analytical method for determining the statistical significance of the estimated value of the STRF pixels has been developed. Application of the method on spike trains collected from individual mouse retinal ganglion cells revealed that the time required to compute the test ranged from a couple of minutes to half a day for different neurons. New method: Here, this method is accelerated by using the Normal approximation to the null distribution of STRF pixel estimates. Results: The significance threshold and computation time obtained under the approximate distribution are examined systematically as a function of various input spike trains collected from individual mouse retinal ganglion cells. Comparison with existing methods: The accuracy and the time saved by the use of the approximate distribution are examined in comparison with the exact distribution. Conclusions: For the real data analyzed here, the approximate distribution yields the same significance thresholds as the exact distribution within a much shorter computation time.