Overview of the TRUP pipeline. The schematic diagram on the left panel shows the four major processing steps applied in TRUP. The cartoon on the right panel illustrates an example of detecting a fusion event. White and black colored boxes indicate reads mapped to gene A and to gene B, respectively. In a first step, TRUP aligns the read pairs onto the genome allowing discovery of chimeric alignments (read pair id p2 and p7 in the cartoon) and partial alignments (p1, p3, p6, and p8). To guarantee a sensitive detection of candidate regions containing potential breakpoint, relaxed criteria are adopted to call breakpoints from chimeric/partial alignments, as well as from entirely aligned discordant pairs (p4 and p5). Subsequently, to reach high accuracy, de novo assembly is performed on a candidate region by using the read pairs anchored in this region. Lastly, bona fide breakpoints relative to the genome are identified from the assembled sequences. A fusion candidate is called if it attracts a sufficient number of supporting reads. While the mapping and assembly steps adopt the state-of-the-art algorithms, the breakpoint searching and fusion calling steps are novel (Materials and methods).