Prediction accuracy of demographic and cognitive metrics from different subsets of the Krakencoder’s latent space and observed SC and FC data. Variability of prediction accuracy was assessed through bootstrap resampling (n = 100). Bar height, box and whiskers represent the mean, 50% and 90% bootstrap confidence intervals, respectively. *Significant difference between prediction accuracy from latent space and the observed data (P perm < 10 -4 , two-sided permutation test, 10,000 permutations, FDR-corrected across 18 comparisons). Bal.Acc, balanced accuracy.
