Beadchip arrays

For validation, we used three independently ascertained sample series; our own sample series with gene expression estimated using Illumina beadchip arrays^{7 (link)}; the frontal cortex series from Colantuoni et al., assayed using a custom array^{24 (link)}; and the GTEx dataset that used RNA-Seq but with multiple brain regions and estimated gene expression per gene rather than per transcript. In each case, we used the dataset provided normalized values (e.g. FPKM reads for GTEx) to calculate the association with age (and nominal p value) and gene expression for each probe using the WGCNA package. We then matched genes and compared R values in each series. Where there were multiple potential matches, for example where we had multiple transcripts in our dataset to compare against gene-level estimates in GTEx, we used the transcript with the highest mean expression in our dataset. We did not consider genes in our dataset where multiple transcripts diverged widely, or genes in the array datasets that were not detected.

DNA methylation datasets for PCa were identified in the Gene Expression Omnibus (GEO) database, using keyword phrase “prostate cancer methylation” [51 (link), 52 (link)]. We identified 16 datasets containing DNA methylation data from PCa, and two of them (GSE26126 and GSE76938) were selected for the analysis. The third methylation dataset used in the analysis was from the Cancer Genome Atlas (TCGA). Information about all datasets is displayed in Table 1. The criteria for selection was that the cohorts were of sufficient size (> 100 samples), contained both cancer and normal samples, and used comparable platforms for analysis (Illumina BeadChip arrays). The similarity in platform facilitated more easy comparison of methylation sites between the datasets.