% IMPORTANT: The following is UTF-8 encoded.  This means that in the presence
% of non-ASCII characters, it will not work with BibTeX 0.99 or older.
% Instead, you should use an up-to-date BibTeX implementation like “bibtex8” or
% “biber”.

@INPROCEEDINGS{Bouss:1041679,
      author       = {Bouss, Peter and Stella, Alessandra and Palm, Günther and
                      Grün, Sonja},
      title        = {{S}urrogate techniques to evaluate significance of spike
                      patterns},
      reportid     = {FZJ-2025-02384},
      year         = {2022},
      abstract     = {Introduction / MotivationSurrogate spike train data is used
                      to generate the null hypothesis in the context of the
                      significance analysis of spike train correlations and
                      spatio-temporal spike patterns. In our work, we compare five
                      different surrogate techniques against the classical
                      technique called Uniform Dithering (UD [1]). In particular,
                      we discuss the use of the surrogates to generate the
                      null-hypothesis distribution in the statistical test of the
                      SPADE method (Spike PAttern Detection and Evaluation [2,3]),
                      which detects spatio-temporal spike patterns in parallel
                      spike trains. In SPADE, both spike trains and surrogate
                      realizations are discretized into 0-1 sequences
                      (binarization) before the pattern detection. We discover
                      that binarized surrogates have a lower spike count than the
                      original data, due to the change in the surrogate’s
                      inter-spike interval (ISI) distribution caused by the UD
                      algorithm. The spike count mismatch between the original
                      data and the surrogates is predominant in the case of high
                      firing rates, spiking regularity, and presence of a dead
                      time (minimal temporal distance between spikes typically
                      induced by spike sorting). We prove that spike count
                      reduction leads to false positive (FP) detection, motivating
                      us to explore alternative surrogate techniques to
                      UD.MethodsSurrogate Techniques (Figure 1)Uniform Dithering
                      (UD)[1] displaces each spike individually according to a
                      uniform distribution.To account for the dead-time present in
                      experimental data, Uniform Dithering with Dead-Time (UDD)
                      limits the displacement of each spike such that the
                      dead-times are conserved.(Joint)-ISI dithering [4] displaces
                      each spike individually preserving the (joint-)ISI
                      distribution.Window Shuffling shuffles binarized spike
                      trains inside of short windows. Trial Shifting consists in
                      shifting entire segments of a spike train according to a
                      uniform distribution, independently trial by trial, and
                      neuron by neuron [1, 5].SPADESPADE detects spike patterns at
                      a millisecond resolution, allowing for temporal delays
                      between the spikes. The spike trains are first discretized
                      and patterns are then mined and counted. To assess the
                      significance of these patterns, the same procedure is
                      performed on multiple realizations of surrogate spike
                      trains, resulting in a p-value spectrum [3]. Significant
                      patterns have p-values lower than the (corrected)
                      significance threshold.Artificial data generationIn order to
                      evaluate the effect of the different surrogates on SPADE, we
                      create artificial spike trains modeled according to the
                      statistics of experimental data from the pre-/motor cortex
                      of macaque monkeys [6], including non-stationary firing rate
                      profiles. The dead time and regularity of the data are
                      modeled by simulating Poisson processes with dead-time (PPD)
                      and Gamma spike trains, respectively. In particular, using a
                      Gamma process the coefficient of variation can be adjusted
                      explicitly, thus allowing for the generation of regular and
                      bursty processes.Figure 1. Surrogate Techniques. A) Uniform
                      Dithering (UD) displaces each spike according to a uniform
                      distribution centered on the spike. B) Uniform Dithering
                      with dead-time (UDD) is based on uniform dithering, but
                      spikes are constrained not to be closer to each other than a
                      dead-time C) Joint Inter-Spike Interval Dithering (JISI-D)
                      displaces each spike according to the J-ISI distribution of
                      the neuron. D) Inter-Spike Interval Dithering (ISI-D)
                      displaces each spike according to the ISI distribution of
                      the neuron. E) Trial Shifting (TR-SHIFT) shifts each trial
                      according to a uniform distribution. F) Window Shuffling
                      (WIN-SHUFF) shuffles binned spike trains within
                      windows.Results and DiscussionWe observe that UD surrogates
                      modify the ISI distribution of PPD and Gamma spike trains
                      (approximately) into an exponential distribution. As a
                      consequence, spike counts are reduced after binarization. By
                      applying SPADE on the artificially generated data, we
                      observe a high number of false positives when employing UD
                      (see Figure 2). Thus, we conclude that UD is not a suitable
                      surrogate technique for spike train data that either
                      contains a dead-time or is regular. The alternative
                      surrogate methods, instead, yield a consistently low number
                      of false-positive patterns; between 8 and 15 FPs in 48
                      analyzed datasets (except for UDD on Gamma spike trains). In
                      conclusion, since trial shifting is the simplest method
                      among the best-performing ones, we recommend it as the
                      method of choice.Figure 2. False-positive patterns in
                      artificial data. The figure shows the number of false
                      positives (FPs) detected per surrogate techniques
                      (color-coded) normalized over the 48 data sets analyzed
                      (y-axis), left for PPD and right for Gamma process data
                      analyses. Numbers at the x-axis indicate the total number of
                      FPs over all data sets per surrogate technique.Key
                      Wordsspike patterns, surrogate techniques, motor
                      cortexReferences[1] Louis S, Borgelt C, Grün S. Generation
                      and Selection of Surrogate Methods for Correlation Analysis.
                      In: Rotter S, Grün S, editors. Analysis of Parallel Spike
                      Trains. Berlin: Springer; 2010. p. 359–382.[2] Quaglio P,
                      Yegenoglu A, Torre E, Endres DM, Grün S. Detection and
                      evaluation of spatio-temporal spike patterns in massively
                      parallel spike train data with SPADE. Frontiers in
                      computational neuroscience. 2017;11:41.[3] Stella A, Quaglio
                      P, Torre E, Grün S. 3d-SPADE: Significance evaluation of
                      spatio-temporal patterns of various temporal extents.
                      Biosystems. 2019;185:104022.
                      doi:10.1016/j.biosystems.2019.104022. [4] Gerstein GL.
                      Searching for significance in spatio-temporal firing
                      patterns. Acta Neurobiol Exp. 2004;64:203–207.[5] Pipa,
                      G., Wheeler, D. W., Singer, W., and Nikolie, D. (2008).
                      NeuroXidence: reliable and efficient analysis of an excess
                      or deficiency of joint-spike events. J. Comput. Neurosci.
                      25, 64–88. doi: 10.1007/s10827-007-0065-3[6] Brochier T,
                      Zehl L, Hao Y, Duret M, Sprenger J, Denker M, et al.
                      Massively parallel recordings in macaque motor cortex during
                      an instructed delayed reach-to-grasp task. Scientific Data.
                      2018;5:180055. doi:10.1038/sdata.2018.55.},
      month         = {Feb},
      date          = {2022-02-22},
      organization  = {6th HBP Student Conference on
                       Interdisciplinary Brain Research,
                       Innsbruck (Austria), 22 Feb 2022 - 25
                       Feb 2022},
      subtyp        = {After Call},
      cin          = {INM-6 / IAS-6},
      cid          = {I:(DE-Juel1)INM-6-20090406 / I:(DE-Juel1)IAS-6-20130828},
      pnm          = {5232 - Computational Principles (POF4-523) / 5234 -
                      Emerging NC Architectures (POF4-523) / HBP SGA2 - Human
                      Brain Project Specific Grant Agreement 2 (785907) / HBP SGA3
                      - Human Brain Project Specific Grant Agreement 3 (945539) /
                      HAF - Helmholtz Analytics Framework (ZT-I-0003) / GRK 2416 -
                      GRK 2416: MultiSenses-MultiScales: Neue Ansätze zur
                      Aufklärung neuronaler multisensorischer Integration
                      (368482240)},
      pid          = {G:(DE-HGF)POF4-5232 / G:(DE-HGF)POF4-5234 /
                      G:(EU-Grant)785907 / G:(EU-Grant)945539 /
                      G:(DE-HGF)ZT-I-0003 / G:(GEPRIS)368482240},
      typ          = {PUB:(DE-HGF)24},
      url          = {https://juser.fz-juelich.de/record/1041679},
}