Methodological scheme.

(A) Synthetic spike trains of 50 neurons during 100-time bins. Four ensembles were artificially generated (see Methods for details). (B) PCA is performed on all the population vectors of more than two active neurons, we used in this case three. For visualization purposes, we plot each of them as a point in the 3D space spanned by the first three principal components (PC). The points tend to group in clusters. (C) The Euclidean distance between all the points on the PC coordinates is computed. The density of each data point ρ and its distance to the next data point with a higher density of δ is obtained. The points above the threshold (blue dashed line, see Methods for details) are considered cluster centroids. (D) All points are assigned to their closest centroid, building up the clusters of population vectors. (E) The correlation matrix between neurons and clusters, corr(n, e) is represented by the symmetric red-blue color map. (F) A threshold matrix is computed to define the significance of corr(n, e) values. (G) Core-cells are detected based on corr(n, e) values that exceed the threshold (red if core, white otherwise). (H) The inner-cluster mean correlation is compared against a threshold of non-core cell correlation, discarding any cluster below this threshold (cluster 1, in this case). This final filtering yields the set of ensembles. (I) Each population vector shown in A is coloured according to the cluster to which they belong. Black vectors were discarded by the minimum number of spikes or the threshold rejection criteria. (J) The detected ensemble sequence is represented as an integer sequence, where the color corresponds to the population vector color in I. Note that the rejection criteria discarded cluster 1, so only ensembles 2–5 are shown, and black circles denote non-clustered population vectors.