Low SV Call Output from CuteSV

[bioteksampath]

Hi,

Do you have any idea why CuteSV reports a significantly lower number of SV calls compared to other tools?

It doesn’t seem to be an unusual case. I mapped my ~4 Gb genome using Minimap2 with RG tags added, and called SVs using three different tools. However, CuteSV returned very few variants.

I used the default depth cutoff of 10 for all callers. Here’s a summary of the commands and call counts:

SV calling with Sniffles2 (deth auto)

sniffles -i ${id}.sorted_dedup.rg.bam -v ${id}.ng_1snf2.vcfraw -t $NSLOTS --minsvlen $min_SV_size
grep "PASS" ${id}.ng_1snf2.vcfraw > ${id}.ng_1snf2_pass.vcf

SV calling with CuteSV (depth 10) also tried with depth -3

cuteSV -t $NSLOTS -l $min_SV_size ${id}.sorted_dedup.rg.bam "$refgenome" ${id}.ng_2cSV_raw.vcf SVout_${id}

SV calling with SVIM

svim alignment SVout_${id}/ ${id}.sorted_dedup.rg.bam "$refgenome" --min_sv_size $min_SV_size --minimum_depth 10
ln -s SVout_${id}/variants.vcf ${id}.ng_3svim_raw.vcf

[Lupphes]

I have a similar observation. I do not know what the issue is.

[Meltpinkg]

Hello, @bioteksampath

Sorry for my late reply. Thank you for bringing this issue to my attention. I’m currently still working to resolve it and it seems that cuteSV occurred some problems and missed massive SVs. If possible, could you kindly provide a small portion of the BAM file—for example, a specific chromosomal region extracted using samtools (e.g., samtools view -b input.bam chr:start-end > subset.bam)? This targeted dataset would greatly aid in identifying and addressing the error.

Thank you again for your time and support. I will continue to investigate this issue, and please let me know if you need any additional details from my side.

Best,
Shuqi

[bioteksampath]

Thanks for your response, attached here is the sample bam from chr1, let me know if further info needed.

subset_bam.txt

[Meltpinkg]

Hello, @bioteksampath

Thanks for sharing the dataset. The problem comes from the parameter -q/--min_mapq in cuteSV, which is set to 20 by default. This means cuteSV only considers the alignments with mapping quality ≥ 20. In the sample bam above, the alignments have a mapping quality from 1~16, which is all smaller than 20, thus they are all filtered by cuteSV. To solve this, I recommend adjusting -q to a lower value (e.g., -q 5) that better matches the quality distribution of your data. This should allow cuteSV to process the alignments as expected.
Hope it will help!

Best,
Shuqi

[Lupphes]

Hello,

Thank you for looking into it. Shouldn't then the cuteSV return an empty VCF file if it does not match any variants with that quality?

[Meltpinkg]

Hello, @Lupphes

This observation may involve a design in cuteSV's handling of supplementary alignments for SV detection: When a structural variant is represented through supplementary alignments, while the primary alignment must still meet the strict --min_mapq threshold, we apply more lenient criteria when parsing the associated supplementary alignments. This explains why you may detect SV signatures from alignment segments that don't themselves meet the MAPQ threshold.

Best,
Shuqi

[bioteksampath]

@Meltpinkg - thanks for your resposne, your approach improved a bit (5k-8k SVs with -q 5 (8k) or -q 3 (9k) ) but not substantially- still missing lot of SV. Wondering do you have any other suggestions?

[Meltpinkg]

Hello, @bioteksampath

This observation appears unusual and warrants further investigation. I have two ideas. First, you can check if there are any error or warning messages in cuteSV's running logs that might explain the discrepancy. Second, I recommand to change the minimum mapping quality to 1, since according to the shared subset bam above, there are 21 of 34 reads with mapping quality equals 1. In this way, we inferred that "-q 3" may filter over 2/3 alignment reads.

In addition, we analyzed the shared subset bam above using cuteSV 2.1.2 with cuteSV test.bam ref.fa test.vcf ./workdir -s 3 -l 50 -q 1, and detected 6 DELs and 10 INSs. You can verify whether your cuteSV results show matching SV counts in this region. If counts differ, you can share both your command and SV detection results for this interval.
If your cuteSV results match ours, you can compare these counts with Sniffles2/SVIM's detection in the same region and provide the differences. (We also ran Sniffles2.2 run with sniffles -i test.bam -v sniffles2.vcf --minsvlen 50, and found no SVs in this interval, which appears particularly unexpected.)

Best,
Shuqi