Fig. 4. Top three hidden subnetworks that underpin structural dependence patterns in social brain differences.

Depicts the uncovered hidden representations that were learned by the autoencoder neural network, involving n = 10,000 UK Biobank participants, with the allocated relevance of each region from the social brain atlas. Red (blue) colors represent the extent of positive (negative) relevances, as depicted in the color bar. The strongest three hidden subnetworks 7, 9, and 15 represented meaningful structural inter-dependencies that highlighted key regions including the TPJ, medial PFC and NAC. For abbreviations and details on social brain regions see Supplementary Table 2. Source data are provided in Supplementary Data 1.