Once the parcellation has been created, parcellated representations of data from each modality can be generated using either the group parcellation or the individual subject parcellations. For the statistical cross-validation, we created parcellated myelin, cortical thickness, task fMRI, and resting state functional connectivity datasets using the semi-automated multimodal group parcellation (see Supplementary Methods 7.1). For myelin and cortical thickness, we simply averaged the values of the dense individual subject maps within each area. For task fMRI, we averaged the time series within each area prior to computing task statistics (to benefit from the CNR improvements of parcellation demonstrated in Fig. 4e). For the same reason, we averaged resting state time series within each parcel prior to computing functional connectivity to form a parcellated functional connectome.
For each pair of areas that shared a border in the parcellation, we computed a paired samples two-tailed t-test across subjects on these parcellated data for each feature (ignoring tests that involved the diagonal in the resting state parcellated functional connectome). We thresholded these tests at the Bonferroni-corrected significance level of P < 9 × 10−8 (number of area pairs across both hemispheres (1,050) × number of features (266) × number of tails (2) × 0.05) and an effect size threshold of Cohen’s d > 1. We grouped the features into 4 independent categories (cortical thickness, myelin, task fMRI, and resting state fMRI) to determine for each area pair whether it showed robust and statistically significant differences across multiple modalities. For more details, see Supplementary Methods 7.2.