Skip to content

Measurement of uncertainty of CNV breakpoints

The DRAGEN CNV caller algorithm uses deviation in expected read depth in genomic regions (~1,000bp windows) to identify gains or losses in genomic content.

Due to the nature of this algorithm, the breakpoints reported by the pipeline are not an accurate representation of exact CNV breakpoints.

Our comparison of DRAGEN CNV breakpoints with the breakpoints obtained from more precise approaches (Manta software, which uses information from split reads and unproperly paired sequencing reads) demonstrated:

  • For CNVs with the breakpoints in well mappable regions (i.e. where sequencing reads can be unambiguously aligned to genome reference), 86.3% of breakpoints determined by DRAGEN CNV are within 1000bp from the breakpoint determined by split or unproperly paired reads, and 94.3% are within 2000bp.1

  • For CNVs where the breakpoint falls within a difficult to map region, e.g. a segmental duplication, the breakpoints cannot be unambiguously determined using short read sequencing technology, and the uncertainty depends on the size of the difficult to map region.

  1. CNV events which are fragmented (i.e. multiple DRAGEN calls overlap with a continuous region that is lost or gained in the sample) will decrease measurement of uncertainty metrics. Fragmentation in DRAGEN calls can occur as a result of CNV events overlapping regions of the genome that are excluded from CNV calling, e.g. regions where read alignment is difficult. ↩