Deep learning identifies partially overlapping subnetworks in the human social brain

Area of Science:

• Neuroscience
• Social Neuroscience
• Neuroimaging

Background:

• Human social interaction is complex.
• Previous research often localized social functions to specific brain circuits.
• Large-scale population datasets for bottom-up brain variation studies are scarce.

Purpose of the Study:

• To deconstruct social brain morphology into fundamental components using a large dataset.
• To identify population-level patterns of structural co-variation in the social brain.
• To relate these structural patterns to psychological concepts of social functioning.

Main Methods:

• Utilized deep learning (autoencoder neural networks) on structural neuroimaging data from ~10,000 UK Biobank participants.
• Performed a systematic deconstruction of social brain morphology.
• Explored coherent representations of structural co-variation within a social brain atlas.

Main Results:

• Identified essential subnetworks representing structural relationships between social brain regions, with key areas including nucleus accumbens, medial prefrontal cortex, and temporoparietal junction.
• Some subnetworks predicted general social traits.
• Other subnetworks were associated with specific social functioning aspects, like social isolation.

Conclusions:

• Spatially overlapping subsystems within the social brain are likely related to interindividual differences in social life.
• This population-level evidence supports a network-based understanding of social brain function.
• Deep learning approaches can reveal fundamental structural relationships in large neuroimaging datasets.