Mapping and quantification of genes, exons and transcripts: The raw reads were trimmed with Trimmomatic (v0.36)10 and then mapped to human reference genome GRCh38.v24 (ftp://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_24/GRCh38.primary_assembly.genome.fa.gz) using STAR (v2.5.3a)11. The BAM files that were generated contain the mapped paired-end reads, including those spanning splice junctions. Following read alignment, expression quantification was performed at the transcript isoform level using RSEM (v1.3.0)12 and then summarized at the gene level. Gene quantifications correspond to GENCODE v27 (ftp://ftp.ebi.ac.uk/pub/databases/gencode/Gencode_human/release_27/gencode.v27.annotation.gtf.gz). Quality control metrics were reported with RNA-SeqQC (v1.1.7)13. All analysis used log2 counts per million (CPM) following TMM normalization14 implemented in edgeR (v3.22.5)15. Correction for GC content bias was performed with cqn (v1.26.0)16. Genes with over 1 CPM in at least 50% of the experiments were retained.
