De novo transcriptome assembly was performed on RNA-seq data with Cufflinks v1.0.1 using the upper quartile normalization (-N) and fragment bias correction (-b) options. This transcript assembly was performed using reads that were prealigned to hg18 using TopHat as described above. In cases where multiple datasets of the same library type from the same tissue were available, these datasets were combined to increase read depth for de novo assembly (see Table S2). For paired end read datasets, only properly paired and singleton reads as defined by SAMTools were used.
