Gast ::
| Hauptseite > Publikationsdatenbank > Spatio Temporal Spike Pattern Detection in Massively Parallel Spike Trains using Formal Concept Analysis |
| Hauptseite > Publikationsdatenbank > Spatio Temporal Spike Pattern Detection in Massively Parallel Spike Trains using Formal Concept Analysis |
| Poster (After Call) | FZJ-2016-03745 |
; ; ; ;
2016
Abstract: Cortical neurons form a highly interwoven network. The observations that i) spike time coordination atmillisecond precision shapes synaptic efficacy [1], ii) neurons emit a spike more reliably upon synchronousthan asynchronous input [2], and iii) synchronous input may result from pre-synaptic spikes emitted at dif-ferent times but arriving simultaneously at the post-synaptic site, suggest that millisecond-precise temporalsequences of spikes emitted from several neurons at successive times may play a role in cortical process-ing. Today modern electrophysiological techniques enable to record hundreds of neurons simultaneouslyincreasing the chances to observe neurons involved in temporal spike sequences, or “spatio-temporal pat-terns” (STPs). Here we propose a method to investigate the presence of STPs in such massively parallelspike train (MPST) data.The analysis of MPST data faces computational and statistical challenges due to the immense numberof possible patterns to evaluate. Existing methods to analyze correlations in MPST data [3-5] typicallyfocus on spike synchrony, a special type of STP. Our suggested method analyzes and extracts STPs ina more general sense by using Formal Concept Analysis. A temporal window is slid along the data andSTPs within each window are extracted. Patterns involved in as many windows and neurons as possible(“formal concepts” [6], also called closed item sets in the data mining community [7]) are identified. MPSTdata typically contain a large number of chance patterns, simply due to the background activity of manyneurons. To further identify non-chance STPs, we apply either a stability analysis [8] of the patterns, or ananalysis of their statistical significance analogous to that proposed in [5].We test our method on ground truth MPST data generated by stochastic simulations. A variety ofparameters affects the performance of the method (in terms of false positive and false negative detections),such as the number of STP occurrences, the number of neurons involved in each pattern and variousfeatures of the background activity which are typical of real data, like firing rate variability over time andacross neurons, and inter-spike interval regularity. We demonstrate the robustness of our method to theseparameters by simulating a number of scenarios which replicate such features. Our results show that themethod is suited for the analysis of STPs in massively parallel spike trains thereby offering the possibilityto relate such patterns to behavior and show their computational relevance.
|
The record appears in these collections: |