Figure 3:

Comparing runtime for Vulcan, NGMLR, and minimap2. The time was measured in terms of CPU time for all programs. A: Vulcan achieves an approximately linear acceleration with the increase of the cut-off percentile. With a 90% percentile cut-off, Vulcan only takes approximately one-fourth of NGMLR's CPU time. B: The majority of Vulcan's CPU time is spent in running NGMLR on the subset of reads, leading to an improvement of their alignments. C: In 90% percentile cut-off, NGMLR only re-aligns 10% of the reads, leading to time usage similar to that of minimap2.