correlated with transcriptional profiles of all genes for each iteration of the null model. Then a null category score for the prior gene set was obtained. This procedure was repeated 10,000 times yielding a null distribution of category scores. The empirical category score of the prior set was tested against the null distribution, and a one-sided p-value was obtained and corrected for multiple comparisons across prior gene sets using the Benjamini-Hochberg FDR method [41], with FDR-corrected p<0.05 deemed as significant. The GCEA was performed using R packages clusterProfiler [42] and DOES [43] with customized codes.
