Figure 3. . Classifying siblings versus unrelated populations when other sibling pairs of the same family are included in the training set (sample of youths). (A) The distributions of out-of-sample accuracy (N = 1,000) when predicting siblings and unrelated participants (blue) versus the null hypothesis (red), which was obtained by running the same classifiers (N = 1,000) with the same features but randomizing the labels (i.e., sibling or unrelated). Each distribution highlights the percentiles 2.5 and 97.5 with a thin line. Thick lines are used to highlight the percentiles 25 and 75, while the central markers are used to show the mean values. (B) The out-of-sample accuracies when predicting only siblings. (C) The out-of-sample accuracies when predicting only unrelated participants. (D) The consensus’ ROIs per functional network (as defined by Gordon et al., 2014) used in the classifier. (E) The location of such ROIs on the surface of the brain.