Fig. 1.

Summary of Methodological Approach. (A) We aimed to define broad factors describing a child’s environment, cognition, and brain network organization by reducing high-dimensional data with three unsupervised machine-learning methods: longitudinal bifactor analysis, principal components analysis, and non-negative matrix factorization. (B) To investigate whether children’s general exposome is encoded in their multivariate patterns of PFN topography, we trained ridge regression models using two-fold cross-validation across our matched discovery and replication samples. To ensure that model performance was not influenced by the choice of split, we also performed repeated random cross-validation using multiple random splits. (C) To investigate associations among the exposome, PFN topography, and cognition, we trained three models: Model 1 (“Exposome”) used only a participant’s general exposome score; Model 2 (“PFN Topography”) used each participant’s multivariate pattern of PFN topography; and Model 3 (“Exposome + PFN Topography”) used both exposome scores and PFN topography.