Fig. 2

Performance of SQUID and other methods on simulation data. a, b Different numbers of SVs (200, 500, and 800 SVs) are simulated in each dataset. Each simulated read is aligned with the aligners (a) STAR and (b) SpeedSeq. If the method allows for user-defined minimum read support for prediction, we vary the threshold from 3 to 9, and plot a sensitivity–precision curve (SQUID and LUMPY), otherwise it is shown as a single point. c, d Performance of SQUID under different RNA-seq experimental parameter combinations (read lengths of 51, 76, and 100 bp combined with fragment lengths of 250 and 350). A longer read length increases both the precision and sensitivity of SQUID. A longer fragment length slightly decreases SQUID’s performance. A short read length with a long fragment length leads to the worst precision and sensitivity. SV structural variant