Extremely high proportion of BND?

[Asadi7163]

I am running Severus on ONT long-read tumor–normal WGS data from a mm39 tumor model. The pipeline completes successfully and produces VCFs, but I am seeing that the overwhelming majority of somatic SVs are labeled as BND rather than DEL/DUP/INV/INS. Specifically, in my last two runs (1 without a VNTR-BED and 1 with) I saw that 99% of the SV's are labeled as BND.

Both tumor and normal BAMs are coordinate-sorted and indexed, aligned with minimap2, and I provided the mm39 TRF VNTR-BED to my second run. The data are not haplotype-tagged.

My questions:
1. Is this expected behavior for cancer genomes and/or ONT data?
2. Does this indicate that the SV graph is too complex to resolve simple events?
3. Are there recommended filters or post-processing steps to convert BNDs into more interpretable SVs?
4. Would adding phasing (e.g., WhatsHap) or increasing coverage change BND resolution?

I would appreciate any guidance on how to interpret or reduce the BND calls. Thank you!

[aysegokce]

Hello @Asadi7163,
I strongly recommend running Severus with a VNTR annotation file.

Severus uses conservative criteria for classifying events as DEL/DUP/INV. If there is insufficient evidence for a deletion (e.g. no clear coverage drop), the event is classified as a BND rather than being force-labeled as a DEL.

SV calling and svtype assignment are independent of phasing.

We have observed similar behavior in mouse data. The mouse reference genome is lower quality than the human reference, leading to reduced alignment accuracy. This can result in weaker coverage signals and an increased number of events being reported as BNDs.

To investigate further, could you please provide:

1. The Severus command line and log file
2. Confirmation that these are BND calls (not sBNDs)
3. The total number of SVs in the VCF

Best
Ayse