%0 Conference Paper
%A Bouss, Peter
%A Stella, Alessandra
%A Palm, Günther
%A Grün, Sonja
%T Surrogate methods for robust significance evaluation of spike patterns in non-Poisson data
%M FZJ-2021-04524
%D 2021
%X Spatio-temporal spike patterns were suggested as indications of active cell assemblies. We developed the SPADE method [1-3] to detect significant spatio-temporal patterns (STPs) with millisecond accuracy. STPs are defined as repeating spike patterns across neurons with potential temporal delays between the spikes. The significance of STPs is derived by comparison to the null hypothesis of independence implemented by surrogate data. SPADE first discretizes the spike trains into bins of a few ms and clips bins with more than 1 spike to 1. The binarized spike trains are examined for STPs by counting repeated patterns using frequent itemset mining. The significance of STPs is evaluated by comparison to pattern counts derived from surrogate data, i.e., modifications of the original data intended to destroy potential spike correlation but under conservation of the firing rate profiles. To avoid false results, surrogate data are required to retain the statistical properties of the original data as close as possible. A classically chosen surrogate technique is Uniform Dithering (UD), which displaces each spike independently according to a uniform distribution. We find that UD surrogates applied to our data (motor cortex) contain fewer spikes than the original data. As a consequence, fewer patterns are expected and, thus, false positives may be generated. We identified as the reason for this spike reduction a change of the ISI distribution: UD surrogates are more Poisson-like than the original data which are in tendency more regular. Thus UD destroys a potential dead time and, therefore, spikes are clipped away.To overcome this problem, we studied several surrogate techniques, in particular methods that consider the ISI distribution, i.e., a modification of UD preserving the dead time, (UDD), (joint-)ISI dithering, trial shifting [4]. Another ansatz is a surrogate that shuffles bins of already discretized spike trains within a small window. We examined the surrogates for spike loss, ISI distribution, auto-correlation, and false positives when applied to different ground truth data sets. These are stationary point process models but also non-stationary point processes mimicking the statistical features of the experimental data. It turned out that trial-shuffling [4] best preserves the features of the original data and generates few false positives; we used it then for application to real data.References: [1] Torre et al (2016) DOI:10.1523/JNEUROSCI.4375-15.2016. [2] Quaglio et al. (2017). DOI:10.3389/fncom.2017.00041. [3] Stella et al. (2019). DOI:10.1016/j.biosystems.2019.104022. [4] Pipa et al. (2008) DOI: 10.1007/s10827-007-0065-3.
%B Neuroscience 2021 - 50th Annual Meeting
%C 8 Nov 2021 - 11 Nov 2021, Online (USA)
Y2 8 Nov 2021 - 11 Nov 2021
M2 Online, USA
%F PUB:(DE-HGF)24
%9 Poster
%U https://juser.fz-juelich.de/record/902742