The brain network underlying social participation: a multimodal, data-driven investigation

Identifying brain phenotypes influencing social participation may help understand social deficits in psychiatric disorders. Previous research shows methodological inconsistencies, lacking consensus on which brain regions are crucial. Data-driven variable selection may overcome this, facilitating unbiased replication and discovery of social brain regions. We compare data-driven selection to literature-identification of brain regions in explaining social participation variation. In 37,576 UK Biobank participants (mean age 65 ± 8, 53% female) with structural and functional neuroimaging data, social participation (range 0-10) was derived by combining leisure activity participation and friend/family visits. First, literature review identified a subset of brain regions previously associated with social measures. Secondly, recursive feature elimination selected a subset of imaging-derived phenotypes in 25% (n = 9,394) of the sample. Hierarchical regression in the remaining 75% (n = 28,152) compared whether data-selected or literature-identified brain phenotype-sets explained more variance in social participation. Individual p-values were corrected for multiple comparisons using the false discovery rate. Recursive feature elimination selected 198 imaging-derived phenotypes. Data-selected imaging-derived-phenotypes explained more variance in social participation (1.31%) than literature-identified (0.84%, F = 3.17, p < 0.0001). Seventeen imaging-derived-phenotypes were associated with social participation including mid-posterior-cingulate, inferior-frontal/orbital and insular thickness, and functional connectivity between pericentral with medial frontoparietal and cerebellar networks. Multi-modal brain imaging-derived phenotypes can predict small but significant variation in social participation. We confirmed previously identified social brain associations of pericentral and medial frontoparietal, and orbital regions while also implicating novel relationships with the insula, acoustic radiation, and lateral frontoparietal networks. This highlights the value of data-driven approaches in solidifying social brain regional involvement, outperforming literature-based methods, and revealing previously undetected relationships.