Brainbehavior prediction using pairwise statistics as predictors and five ICA-derived cognitive-behavioral components as the outcome 55 . Kernel ridge regression is performed under a nested tenfold cross-validation setting. The heatmap colors display the mean Pearson’s correlation between the empirical and predicted behavior scores across the test folds. The colorbar covers both negative values (0th percentile to zero, in blue) and positive values (zero to 97.5th percentile, in red). Sensitivity analyses, using alternative machine learning algorithms (kernel ridge regression with a cosine kernel, linear ridge regression and LASSO regression) are shown in Supplementary Figs. 8 10 . Pos, positive; Neg, negative.
