Source-linked AI summary
Monitoring spike train synchrony
Thomas Kreuz, Daniel Chicharro, Conor Houghton, Ralph G Andrzejak, Florian Mormann
TL;DR
The paper addresses unreliable instantaneous values in the original SPIKE-distance, the lack of a causal real-time measure, and limited tools for time-resolved synchrony in continuous data. It revises the distance, defines a past-only variant, and applies both to spike-train and continuous-data analyses. The methods track population synchrony and instantaneous clustering, with applicability also extended to continuous data such as EEG.
Problem
The original SPIKE-distance can produce spuriously high instantaneous values for event-like firing, its use of future spikes prevents causal real-time calculation, and time-resolved synchrony measures for continuous data are lacking.
Source: §1 · p. 2, §3.3 · p. 10Method
The paper revises SPIKE-distance using nearest-spike time differences, defines a past-only causal variant, and applies the distances to event sequences extracted from continuous data.
Source: §1 · p. 2, §3.3 · p. 11Results
The revised and real-time distances track group synchronization and instantaneous clustering, and the paper extends both variants to continuous data such as EEG.
Source: §1 · p. 2, §4 · pp. 11–12Takeaways & Limitations
Selective time-interval averaging can retain evolving dissimilarity patterns that a single population value may obscure.
Source: §4 · pp. 12–13When full spike trains are known, the regular distance is locally more reliable; without future spikes, the real-time variant can show high values during reliable events that are not informative.
Source: §4 · p. 12
Abstract
from arXiv · showhide
Recently, the SPIKE-distance has been proposed as a parameter-free and time-scale independent measure of spike train synchrony. This measure is time-resolved since it relies on instantaneous estimates of spike train dissimilarity. However, its original definition led to spuriously high instantaneous values for event-like firing patterns. Here we present a substantial improvement of this measure which eliminates this shortcoming. The reliability gained allows us to track changes in instantaneous clustering, i.e., time-localized patterns of (dis)similarity among multiple spike trains. Additional new features include selective and triggered temporal averaging as well as the instantaneous comparison of spike train groups. In a second step, a causal SPIKE-distance is defined such that the instantaneous values of dissimilarity rely on past information only so that time-resolved spike train synchrony can be estimated in real-time. We demonstrate that these methods are capable of extracting valuable information from field data by monitoring the synchrony between neuronal spike trains during an epileptic seizure. Finally, the applicability of both the regular and the real-time SPIKE-distance to continuous data is illustrated on model electroencephalographic (EEG) recordings.
1. Introduction
Parameter-free SPIKE-distance complements firing-rate-based dissimilarity with spike-timing sensitivity, while time-resolved analysis supports questions requiring fine temporal precision. The introduction presents an improved distance, a causal real-time variant, and applications to seizure and continuous EEG data.
1. Introduction
- The revised SPIKE-distance removes spuriously high instantaneous values for event-like firing patterns by comparing nearest-spike time differences.The original measure could give unreliable instantaneous values even though temporal averages and moving averages reflected expected results and long-term trends.Source: §1 · p. 2
- The causal SPIKE-distance uses past information only, enabling instantaneous dissimilarity estimates in real time.The original instantaneous measures also use following spikes, which prevents real-time calculation.Source: §1 · p. 2
- The revised measures track population synchronization and changes in instantaneous clustering, or evolving dissimilarity patterns among multiple spike trains.The introduction describes validation on artificial spike trains.Source: §1 · p. 2
- The methods add selective and triggered temporal averaging and comparisons between spike-train groups.The introduction also describes applying the methods to single- and multi-unit activity around an epileptic seizure.Source: §1 · p. 2
- The new distances are presented as suitable for characterizing neuronal firing patterns involved in seizure generation, propagation, and termination.The application analyzes spike trains from an epilepsy patient over an interval including a seizure.Source: §1 · p. 2
- For continuous EEG data, the authors transform time series into discrete data and use both SPIKE-distance variants to measure time-resolved dissimilarity.The example uses EEG whose synchronization was modified post hoc by linear mixing.Source: §1 · p. 2
2. Methods
The methods define SPIKE-distance through instantaneous pairwise dissimilarity profiles and their temporal averages, then revise the profile to address spurious high values and make a past-only real-time variant. Examples illustrate the revised and causal profiles, alongside practical calculation considerations.
2. Methods
- For multiple trains, pairwise instantaneous profiles are averaged before temporal averaging; profiles are bounded by [0, 1], with zero only for identical trains.The temporal average of a profile defines the distance for the recording interval.Source: §2 · p. 2
2.1. Original definition of the SPIKE-distance
- The original profile normalizes mean preceding and following spike-time differences by mean interspike interval, weighting nearer spikes more and adding auxiliary boundary spikes.Auxiliary spikes at t = 0 and t = T resolve ambiguity at the first and last intervals.Source: §2.1 · p. 3
2.2. Improved definition of the SPIKE-distance
- The improved profile compares each corner spike with its closest counterpart, avoiding mismatched preceding and following spikes that produced spurious high values.The revised weighting is performed separately for each spike train and accounts for relative distances within each train and differences in firing rate.Source: §2.2 · pp. 4–5
- The improved profile tracks noisy-background jitter and increasingly precise events, reaching zero at the perfectly synchronous event at 3900 ms.The constructed examples show the revised profile reflecting event synchrony without the spurious high values of the original profile.Source: Figure 2 · p. 4, §2.2 · p. 5
2.3. Real-time SPIKE-distance
- The real-time variant uses past information only, enabling online calculation; its causal moving average smooths fluctuations, and common non-spiking lowers dissimilarity.The moving average reflects long-term trends in the bivariate example.Source: §2.3 · p. 5
2.4. Practical considerations
- The improved profile is piecewise linear and the causal profile hyperbolic; sequential segments address memory demands, while computational load scales as N^2.Piecewise integration gives an exact distance; sampled time-resolved calculation approximates it for sufficiently small sampling intervals.Source: §2.4 · pp. 6–7
3. Results
The applications show regular and real-time SPIKE-distances tracking time-varying spike-train clusters and seizure-associated synchrony, and extending event-based analysis to continuous EEG.
3.1. Application to artificially generated data: Instantaneous clustering
- Both SPIKE-distances reflect changing cluster associations, including transitions from multiple clusters to random spiking.The examples change cluster membership every 500 ms; instantaneous pairwise matrices capture these changes even when overall synchrony remains rather constant.Source: Figure 5 · p. 7, §3.1 · p. 7
- Regular SPIKE-distance blends past and future behavior, whereas the real-time variant uses past spikes only and can distinguish neurons by whether they have fired.The real-time measure does not yet reflect a cluster change that occurs in the future relative to the instant being assessed.Source: §3.1 · pp. 7–8
- Selective averaging over chosen, even separated, intervals preserves recurring cluster relationships in the combined dissimilarity matrix.Averaging across the full example still identifies four groups that frequently clustered together; groups that co-clustered in two intervals have lower dissimilarity.Source: Figure 6 · p. 8, §3.1 · p. 8
- Reference-spike triggers isolate five coupled trains, while stimulus-amplitude triggers reveal clustering as chirp amplitude decreases.The internally triggered dendrogram groups the five modified trains with the reference train; externally triggered matrices and dendrograms show an emerging cluster among modulated trials.Source: §3.1 · pp. 8–9, Figure 7 · p. 9
- Averaging dissimilarities within manually assigned spike-train subgroups produces block matrices for comparing groups.The subgroup averages are denoted < >G and can also be represented by corresponding dendrograms.Source: Figure 8 · p. 9, §3.1 · p. 9
3.2. Application to single-unit recordings from epilepsy patients
- During one patient's seizure, focal-hemisphere distances were lower, while non-focal neurons retained a stable synchrony pattern across periods.Both measures also showed a pronounced drop at seizure onset, when only subtle changes were discernible in the local field potential.Source: Figure 9 · p. 10, §3.2 · p. 10
3.3. Application to continuous data
- For continuous EEG, local maxima and minima are treated as event sequences, and the two resulting dissimilarity profiles are averaged.Both regular and real-time measures tracked transitions between independent and synchronous signals, though deviations from identical signals were easier to trace than deviations from independent signals.Source: Figure 10 · p. 11, §3.3 · p. 11
4. Discussion
The discussion presents an improved SPIKE-distance, a causal real-time variant, and extensions for continuous data. These tools support time-resolved analysis of changing spike-train similarity, while the real-time variant is less locally reliable when complete spike trains are available.
4. Discussion
- The revised SPIKE-distance removes spuriously high values for event-like firing patterns; the causal variant uses past information, and both variants extend to continuous data.The paper describes continuous-data application after transformation to discrete data.Source: §4 · pp. 11–12
- Improved reliability supports instantaneous clustering analyses, including pairwise dissimilarity matrices or dendrograms, selective and triggered temporal averages, and comparisons across spike-train groups.Averaging can target continuous or non-continuous intervals of differing lengths and can be internally or externally triggered.Source: §4 · pp. 11–12
- The parameter-free, time-resolved measure is envisaged for awake-animal studies and monitoring synchrony during behavior or evolving stimuli without identical repeated trials.The discussion also identifies possible real-time applications in brain-machine interfaces and seizure monitoring.Source: §4 · p. 12
- When complete spike trains are available, the regular SPIKE-distance is locally more reliable; the real-time variant can nevertheless give similar field-data results with half the information.Future spikes resolve high real-time values that reflect local uncertainty rather than spurious dissimilarity.Source: §4 · p. 12
- Both variants can monitor synchrony in continuous data after transformation to discrete data; mixed continuous-discrete analyses are also possible.The discussion suggests combining spike trains with continuous neuronal oscillations as a potential application.Source: §4 · p. 12
- Reducing instantaneous pairwise values to a single average can conceal variation because high and low dissimilarities may cancel, so the representation should suit the analysis.Higher-order statistics such as variance are suggested as another option.Source: §4 · p. 13
- The measure is time-scale adaptive and parameter-free, and its instantaneous values can be averaged over chosen intervals instead of using sliding-window analysis.SPIKE-distance values are typically below 0.5, while equal-rate Poisson trains have an expectation of 0.295 for SPIKE-distance and 0.5 for ISI-distance.Source: §4 · p. 13
- SPIKE-distance tracks synchrony from spike timing rather than local firing-rate covariation, and complements cross-correlation by measuring instantaneous synchrony rather than dependence on time lag.The discussion says time lags should be eliminated before applying SPIKE-distance.Source: §4 · pp. 12, 14
Appendix A. The ISI-distance
The ISI-distance derives instantaneous dissimilarity from the relative interspike intervals of two trains and averages the absolute ratio over time. Because it requires following spikes, it has no causal real-time extension.
Appendix A. The ISI-distance
- The ISI-distance profile uses the instantaneous ratio of the two trains’ interspike intervals, unlike SPIKE-distance, which derives its profile from differences between spike times.Source: Appendix A · p. 14
- The ISI-ratio is zero for identical intervals and approaches −1 or 1 when the first or second train is much faster, respectively.Source: Appendix A · p. 14
- Temporal averaging of the absolute ISI-ratio yields the ISI-distance, but reliance on following spikes prevents a causal real-time extension.Source: Appendix A · p. 14
Appendix B. Field data: Single-unit recordings from epilepsy patients
The epilepsy-patient recordings were obtained independently of this study’s design and processed from continuously sampled micro-wire signals.
Appendix B. Field data: Single-unit recordings from epilepsy patients
- Recordings preceded and were independent of the study design, and conformed to the University of Bonn Medical Institutional Review Board guidelines.Source: Appendix B · p. 14
- Eight high-impedance platinum-iridium micro-wires protruded from each electrode probe, and bipolar montages were amplified with a 128-channel Neuralynx system.Source: Appendix B · p. 14
- Signals were filtered between 0.1 and 9000 Hz and continuously sampled at 32 kHz; spike detection and sorting followed 300–3000 Hz band-pass filtering.Source: Appendix B · p. 14