Deep learning identifies partially overlapping subnetworks in the human social brain

Complex social interplay is a defining property of the human species. In social neuroscience, many experiments have sought to first define and then locate 'perspective taking', 'empathy', and other psychological concepts to specific brain circuits. Seldom, bottom-up studies were conducted to first identify explanatory patterns of brain variation, which are then related to psychological concepts; perhaps due to a lack of large population datasets. In this spirit, we performed a systematic de-construction of social brain morphology into its elementary building blocks, involving similar to 10,000 UK Biobank participants. We explored coherent representations of structural co-variation at population scale within a recent social brain atlas, by translating autoencoder neural networks from deep learning. The learned subnetworks revealed essential patterns of structural relationships between social brain regions, with the nucleus accumbens, medial prefrontal cortex, and temporoparietal junction embedded at the core. Some of the uncovered subnetworks contributed to predicting examined social traits in general, while other subnetworks helped predict specific facets of social functioning, such as the experience of social isolation. As a consequence of our population-level evidence, spatially overlapping subsystems of the social brain probably relate to interindividual differences in everyday social life. Kiesow et al. use deep learning to identify partially overlapping subnetworks in the human social brain at the population level. They also demonstrate that the learned subnetwork representations can be used to predict social traits.

Automated summary

This study used deep learning to analyze a large population dataset from the UK Biobank, revealing relationship patterns among social brain structures, identifying important subnetworks related to social functioning, and demonstrating that these subnetworks can predict social traits.