Extended Data Fig. 5: Clustering analyses populated across all participants for the binary classifiers.

a,b) Pie charts show the percentage of frequency bands that were selected across all participants for positive and negative decoders, respectively. The histograms show the percentage count of each frequency band within each cluster, implying that the low frequency cluster was mainly made up of the theta, alpha and beta bands. The gamma cluster was mainly made up of the high and low gamma bands for both decoder types. c) left and right panels show the populated normalized feature importance and the stability across all 10 participants for the positive decoders(n = 149 and n = 124 for gamma and low-frequency clusters, respectively), with p values obtained by two-sided pairwise ranksum tests at the bottom of each panel. d) represents similar panels as in C for negative decoders (n = 62 and n = 45 for gamma and low-frequency clusters, respectively). e,f) ratio is defined as (number of features in gamma cluster – number of feature in low frequency clyster) /total number of features contributing to both gamma and low frequency clusters (from Fig. 4b,d), positive ratio means the region have more selected features in gamma cluster and negative ratio means the region has more selected features in low-frequency cluster across subjects. INS: insula, VCin = Ventral cingulate, DCin = dorsal cingulate, AMY: amygdala, OFC = orbitofrontal cortex, HPC = hippocampus. We have generated permuted distributions (that is, null distributions) by shuffling (1000000 times) the region label of each feature and recomputing the ratio (gray boxplots). Confidence intervals are based on the t-statistics since the permuted distribution are normally distributed. All real values of the ratio shown in green(E) and orange(F) circles are outside the confidence interval of the permuted distributions. Confidence intervals in panel E are as following: VCin = [0.0908, 0.0917], DCin = [0.0914, 0.092], HPC = [0.0913, 0.0918], AMY = [0.0913, 0.0919], INS & OFC = [0.0914, 0.0919]. Confidence intervals in panel F for VCin = [0.1584, 0.1594], DCin = [0.1584, 0.1593], HPC = [0.1580, 0.1593], AMY = [0.1582, 0.1594], INS & OFC = [0.1586, 0.1592]. In the box plots(C-F) central lines represent the median and the two edges represent 25 and 75 percentiles; whiskers show the most extreme datapoints, and outliers are shown individually (see MATLAB boxplot function).