RNA-Seq reads were aligned to databases of predicted splice junctions and to gene models using TopHat, STAR and BLAT. A number of criteria were used to distinguish true splice junctions from false positives. First, only reads with a minimum of 6nt overhang across a junction were considered. Second, an entropy score (which reflects the number of offset reads) was calculated for each splice junction, and only junctions with entropy score >2 in at least two biological samples were considered. The set of junctions was filtered further to exclude junctions when intron length less than 41 nt, junctions that join parologous genes and non-canonical splice junctions not supported by other experimental support.