We defined silent and active enhancers from ENCODE HeLa-S3, GM12878 and K562 broad peaks (Broad Institute, Bernstein), downloaded from the UCSC ENCODE repository, according to the co-existence of histone modifications H3K4me1, H3K27ac and H3K27me3. Active enhancers were defined as co-localized H3K4me1 and H3K27ac peaks with no H3K27me3 peak, while silent enhancers were considered loci with H3K4me1 and H3K27me3 peaks but no H3K27ac peak. Loci were filtered to be located distant to TSSs (500 bp) and exons (200 bp) of protein-coding genes, multi-exonic non-coding genes and mRNAs (from ENSEMBL, GENCODE (v10), RefSeq and UCSC, downloaded January 12, 2012), and other lncRNAs from a gene-centric set derived from literature40 as well as manually annotated sense-antisense pairs (coding-noncoding and noncoding-noncoding sense-antisense pairs) with 5' EST and cDNA support, and 5' ESTs with no locus protein-coding capacity. Transcriptional differences between active and silent enhancer sets were determined by comparing the average number of FANTOM5 CAGE tag 5’ ends from the same ENCODE cell lines (pooled triplicates) in a window +/− 300 bp around the H3K4me1 peak mid points.
