Strangely high proportion of variants are INDELs

[PeteH]

I am analysing sequencing data (exome and region capture) of ENU-mutagenised mice. The most recent sample is a capture of ~28Mb region where the mutation has been previously mapped to. After running the sample through our pipeline I found that approx. 60% of variants called within the target region were INDELs compared to approx. 30% for similar-ish exome sequencing projects. All samples are sequencing on the Illumina GAII or HiSeq.

Our pipeline is roughly as follows:

Alignment with quality score recalibration with Novoalign -> remove multimapping reads -> duplicate removal with Picard's MarkDuplicates -> Variant calling with mpileup and bcftools

with mpileup and bcftools parameters as follows:

samtools mpileup -q1 -C50 -d10000 -L10000 -ugf $reference $bamfile | bcftools view -bvcg - > $bcffile
bcftools view $bcffile | vcfutils.pl varFilter -D10000 -w0 -W0

I realise there are considerable differences between my region capture sample and my comparison exome samples (e.g. different capture platforms, generally higher coverage for region capture), however this result still has me a little concerned.

Would local realignment (such as that offered in GATK) be advisable to resolve the status of these putative INDELs?
Does this INDEL proportion strike others as too high?

Any suggestions would be welcome.
Pete

[swbarnes2]

Try redoing the mpileup with -B or -E.

What can happen without those options is the BAQ calculations, which are on by default, sometimes see a real SNP, and decides that its not a SNP, it's an error due to an undiagnosed indel, and mpileup will drop the quality of each of those letters down to nothing, and then the SNP caller won't call it due to high quality. I've seen this happen with a few sanger verified SNPs. But it won't do this for indels, so this can lead to a huge number of indels as compared to simple SNPs.

If you turn off those calculations with -B, or modify them with -E, more real simple SNPs may surface, which may give you a more believable ratio.

[clarissaboschi]

indel analysis

Swbarnes2, about your comment, I am analysing indels using mpileup/bcftools. I had problems to run for all chromosomes together, even for one chromosome (the analyses stopped in the middle of one chromosome), and someone gave me the advice to use the option -B.
But in this case do you think I can have many false-positive indels? I am not using the option -E as well. Only some filtration based on quality scores (q and Q).

My command line is
samtools mpileup –q20 –Q20 –AB -ugf reference.fa file.bam | bcftools view -bvcg - > var.raw.bcf
bcftools view var.raw.bcf | vcfutils.pl varFilter -D99999 > var.flt.vcf

But if I will not use -B, I cannot run my analysis...and I have many individuals to run…

Thanks