Fig. 5: Functional connectivity networks derived from brain regions enriched for schizophrenia SNP-heritability distinguished cases and controls.

A Workflow of the fMRI analysis in two independent data sets, detailed in “Method” section. In each fMRI data set, five folds of recursive feature eliminations were performed using the training set. In each run, the model (network) was evaluated in the independent testing set using the area under curve (AUC) of the receiver operator characteristic curve (ROC). B Number of runs for which each region was preserved in the recursive feature elimination in fMRI data set 1 (46 cases and 46 control) and data set 2 (54 cases and 54 controls). The dot indicates the mean runs in each data set for the region (with standard deviation bars per data set); the color indicates the broader areas each region belongs to. The Pearson’s product-moment correlation between the runs in the two data sets was 0.02 (p = 0.90) for all ROIs, but 0.62 (p = 0.03) for the amygdalar and hippocampal regions, and these regions were also more likely to be preserved to later runs (also Fig S9 C, D). (C) Pairwise connections between regions ranked by the feature importance (FI, y-axis) per connection in data set 1. The color indicates connections involving any amygdalar or hippocampal regions (red) or not (teal). D Density plot of the FI in amygdalar or hippocampal connections (red) and other connections (teal) in data set 1. The FI of amygdalar or hippocampal connections are significantly higher than the others (two-sided Welch two sample t-test, p = 2.2e-84). E Connections ranked by FI in data set 2. F Density plot of the FI in data set 2. The FI of amygdalar or hippocampal connections are significantly higher than the others (two-sided Welch two sample t-test, p = 1.4e-9).