Figure 1.

Overview of the PVS_B and the PVS_U algorithms. Ten-fold cross-validation was performed to obtain a PVS_B for each individual. For each fold, mass univariate summary statistics, , were obtained from the training set which contained 90% of the complete sample. Posterior mean effect sizes at each vertex, , were approximated by multiplying the mass univariate beta estimates, , by the inverse of the correlation structure of the brain, D, and a shrinkage factor that accounts for the number of vertices, V, and the total signal of the brain-behavior association. The PVS_B was subsequently calculated for the test set participants by multiplying their imaging phenotype with the . Simulations were conducted at three levels of total explainable signal, six levels of study sample size, and four levels of proportion of non-null vertices, yielding 60 instantiations of simulation conditions with 100 iterations per condition.