Significance of Pindel's "AD" and finding SVs in IGV

[opulcy97]

Hi All,

I am running pindel on some nextgen data for exploring ins, del, rpl, tandem:dup, etc. I have two questions:

1. I can see most of the deletions from pindel produced by the same bam files in IGV. However, I do not see most of the insertions and replacements. Can you please tell me why is that so? Are they false positive produced by pindel? Or, does IGV has a different way of predicting ins/dels/rpls than pindel? I am using all default parameters.

2. I know AD = "Allele depth, how many reads support this allele". Can I filter out SVs based upon AD? I found a real deletion (confirmed by sanger sequencing) predicted by pindel with AD = 8! Whereas, there are some deletions which have far more reads but turned out to be false positives. Why is that so? What happens if pindel shows AD = 0. Does that mean that the prediction is unreliable as there are no reads to support that prediction?

Sorry for so many questions. But I am banging my head for weeks to look for answers over different forums and couldn't get hold of it.

Thanks in advance.

Regards

[nnutter]

This question was also asked on BioStar: http://www.biostars.org/p/75407/

[opulcy97]

Still waiting for a reply. Not like the one from the nutcase.

[KaiYe]

Pindel re-align every read, which does not have perfect match to the reference genome: unmapped, gaped aligned, clipping, with too many mismatches and so on. IGV is just visualizing alignment produced by the aligner so that any indels missed by the primary aligner would not appear in IGV.

you should look at the AD and also the total coverage for germline calling/filtering. a simple way is to calculate the percentage of alt supporting reads. say you see 10 reads support ref and 8 alt, this is a good het call. however if you see 1000 reads support ref, and again 8 for alt, you probably won't consider it as germline het.

If you run Pindel on one sample, you won't see any variants without reads supporting alt. but you will see them when you have multiple samples.

[opulcy97]

Hi KaiYe,

Many thanks for the reply. I didn't assume that pindel supports multi-sample calling. I thought when you specify multiple samples in configuration file, it actually calls them individually! That was really stupid of me. Currently, I am running pindel on single sample and there are 10 samples (not tumor - normal pairs) which have some know genetic phenotypes. Do you think running multiple samples for CNV with pindel will produce more accurate predictions rather than single sample?

Also, what did you mean by "the total coverage for germline calling/filtering"?

Many thanks again.
Regards

PS. Reiterating one of my last questions: What happens if pindel shows AD = 0. Does that mean that the prediction is unreliable as there are no reads to support that prediction?

[KaiYe]

multiple sample calling is certainly better, in terms of sensitivity.

[opulcy97]

Hi KaiYe,

I was just wondering whether it is possible to, somehow, obtain pindel re-aligned bam file from the pindel run.

Many thanks indeed.
Regards

[gzhmat]

Pindel re-align every read, which does not have perfect match to the reference genome: unmapped, gaped aligned, clipping, with too many mismatches and so on. IGV is just visualizing alignment produced by the aligner so that any indels missed by the primary aligner would not appear in IGV.

you should look at the AD and also the total coverage for germline calling/filtering. a simple way is to calculate the percentage of alt supporting reads. say you see 10 reads support ref and 8 alt, this is a good het call. however if you see 1000 reads support ref, and again 8 for alt, you probably won't consider it as germline het.

If you run Pindel on one sample, you won't see any variants without reads supporting alt. but you will see them when you have multiple samples.

Hi KaiYe,
Recently, I am using Pindel 0.2.4t August 13, 2012 version to call indels from paired exome sequencing data.
All the process is working fine but some results are very confusing.
One deletion was detected with GATK shows that at 17 reads support the SV, while Pindel only report 5 reads.
I checked the mapping result with 'samtools tview' and it seems that GATK is correct.
Do you have any suggestions?

[KaiYe]

Hi KaiYe,
Recently, I am using Pindel 0.2.4t August 13, 2012 version to call indels from paired exome sequencing data.
All the process is working fine but some results are very confusing.
One deletion was detected with GATK shows that at 17 reads support the SV, while Pindel only report 5 reads.
I checked the mapping result with 'samtools tview' and it seems that GATK is correct.
Do you have any suggestions?

Pindel is picky in selection of reads as the support evidence. Recent version does a better job in gathering variant supporting reads. please update your version to 0.2.5.