Fig. 4: Single-trial analysis of activity onset for classifier and spectral power traces.

a Schematic explanation of single-trial onset analysis. For every single trial the onset of the highest peak of classifier performance (D-values) or single-trial power trace was determined. Here, two example classifier D-value traces are shown for lead P11 located in area PFt, with onsets indicated by vertical lines. Subsequently, either the average onsets are computed across all leads in each area of interest, after which the differences of the average onsets is determined (‘average of onsets’), or the differences in onset of pairs of leads are compared for each trial, after which the differences are averaged within the areas of interest (‘trial by trial’). Both methods thus resulted in a difference matrix between all areas of interest, which was then ordered on a linear time axis using a multidimensional scaling approach. For power traces (not shown in this example), onset analyses were computed for each frequency independently and time difference distributions were combined across frequencies; b–e temporal ordering results of the average of onsets approach (b, d) and the trial-by-trial approach (c, e) for single-trial classifiers (b, c) and spectra (d, e). Connected filled circles indicate the onset of each area relative to the onset of area BA4 (gray), with blue areas starting before BA4 and yellow areas appearing after BA4. Small dots represent bootstrapping results, with each bootstrap leaving out the data from one recording lead (N = 86). The classifier onset distributions of all leads can be found in Supplementary Fig. 12 and time difference matrices for all four methods are given in Supplementary Fig. 13.