Extended Data Fig. 3: Regional variation in prediction accuracy and baseline reliability.

For the Krakencoder’s fusion predictions made without the parcellation being predicted (fusion-parc, fusionFC-parc, and fusionSC-parc), the prediction accuracy for each region is the Pearson correlation between the predicted and measured edge values for 196 held-out test subjects, averaged across all other regions. Age- and sex-matched individuals of given relatedness, as well as self-retest, are correlated and averaged in the same manner, in order to provide a baseline for comparing the spatial patterns of edge-wise accuracy. Only the left hemisphere is shown, to simplify visualization.