Using the same 20 sets of 80%-20% splits used to evaluate differential expression (see “Cross-validation of differential expression”), we estimated the module reproducibility. We generated modules in the controls and SCZ using the 80% split and then examined the reproducibility of connectivity in the independent 20% replication cohort. The connectivity is estimated based on adjacency matrix using the same power (beta = 6) across all comparisons. The median Pearson correlation of connectivity values among the “discovery” and “replication” cohorts was 0.77 (mean = 0.76, sd = 0.06, 25th percentile = 0.73, 75th percentile = 0.78) and 0.80 (mean = 0.78, sd = 0.07, 25th percentile = 0.70, 75th percentile = 0.84) for cases with SCZ and controls, respectively. This strongly supports the robustness of the gene-gene correlation structure, since this replication process occurs in a completely independent sub-cohort of 20% of the brain samples.
