Strelka output using Clipped bams not the same as unclipped bams

[statgen]

Strelka output using Clipped bams not the same as unclipped bams

[pjvandehaar]

Do you think that clipOverlap clipped some reads that should not have been clipped? Can you find any reads that overlap that position that were clipped?

Do you see whether that deletion was called by Strelka but then Strelka filtered it out?

(Disclaimer: These are speculations, and I don't know what I'm talking about.)

This screenshot shows some of the reads supporting this deletion. Two of the read pairs have a small overlap but not at the region of interest. Why Strelka filter out the deletion for one set of bam files but not the others?

[mktrost]

Can you look at those reads before and after clip overlap?

What do the flags and the cigars look like?
Any change in them?

How about their mates? Did they change?
On Jan 13, 2016 4:44 PM, "artonmarton" [email protected] wrote:

[image: screen shot 2016-01-13 at 1 41 46 pm]
https://cloud.githubusercontent.com/assets/12701400/12308702/83664f8e-b9fb-11e5-9013-3cd55eb08639.png

This screenshot shows some of the reads supporting this deletion. Two of
the read pairs have a small overlap but not at the region of interest. Why
Strelka filter out the deletion for one set of bam files but not the others?

—
Reply to this email directly or view it on GitHub
#20 (comment).

Sorry for delay, I am trying to figure out what is the differences, Where can I check the flags?

There are three reads that are different, two of the reads have lower phred score in the clipped bams and the third read has a CIGAR string as attached in this image.

[mktrost]

Unpaired reads should pass through ClipOverlap unchanged, so those reads
should be identical with or without clipOverlap (unless you used the
--overlapsOnly option, but I doubt you did that.

A CIGAR change is expected for an overlapping pair.

I'm intrigued by the phred quality change. ClipOverlap should not modify
any quality scores. So I find that interesting.
If phreds are recaulculated, it is possible the algorithm will come up with
different phreds even on reads without overlaps.

Did you run any tools between ClipOverlap and Strelka?

I don't know how Strelka works, but lower phred scores could affect what it
detects. So it may be worthwhile to figure out how/why they are changing.

On Wed, Jan 13, 2016 at 6:23 PM, artonmarton [email protected]
wrote:

[image: screen shot 2016-01-13 at 3 23 23 pm]
https://cloud.githubusercontent.com/assets/12701400/12311088/9d61a678-ba09-11e5-98af-a1ea52c5fed2.png
This is one of the reads with lower phred score.

—
Reply to this email directly or view it on GitHub
#20 (comment).

No I didn't run anything else between ClipOverlap and Strelka. I only made index files for the bam files after ClipOverlap.

[pjvandehaar]

In the code you ran in issue #16, were some quality scores changed? Could that be the cause? Are your unmodified bams from before that code, or after it?

(Again, disregard this comment if it's incoherent.)

Please ignore Phred score because that is not the issue and that was my mistake that I did not selected the right basepair. I am back to the first square. :)

The reason behind this issue is still unknown if someone has any idea what can cause this inconsistency in the results of Strelka using clipped and unclipped bams.

[pjvandehaar]

Back to mktrost's question from earlier:
When you look at the reads supporting the missing variant (and maybe their mates)– are there any differences (other than the CIGAR changes that are expected) between the reads before and after clipOverlap?

Is your comment from earlier still correct, or are there other reads that should be looked into?

No the comments are correct. There is no difference but CIGAR for some of the reads.

[mktrost]

I don't know how Strelka determines its results. It does seem odd that it would lose the deletion if the only difference for that position is that 80 bases away from it a read is softclipped. If the deletion was in a clipped position that would be an obvious explanation, but you said that wasn't the case. Maybe Strelka takes into account read length and because the matching region is shorter it thinks it is more likely to be mismapped. Do you know anything about how Strelka works or determines it's results? That might help figure it out. I had asked previously about flag differences in case clipOverlap made a read unmapped due to orientation clipping. Because that could be a reason for losing evidence for the deletion. But it sounds like you checked and that isn't the case. Let me know if I'm misunderstanding your question. Fyi there is a bamUtil diff tool (with --all option) that may be helpful for comparing bams if you need it. Unfortunately it doesn't have an option to limit the diff to a specific region. I will try to add that in the future.

As much as I know about Strelka, it doesn't remove reads with different length. Every read has a different length so that is not an issue.

About the question you asked about flags, I couldn't find a way to check the flags. Should I look at the bam files?

I have extracted the reads with all the informations in this region in two files. Do you think the bamUtil diff tool would work on these files?

I tried something new today, I run Strelka using unclipped-bam for normal
and clipped bam for tumor and vice-versa. Surprisingly when ever I used
un-clipped normal bam I got the correct results. So I am guessing maybe
there is a problem with the clipped normal bam?

On Fri, Jan 15, 2016 at 9:32 PM, mktrost [email protected] wrote:

I don't know how Strelka determines its results. It does seem odd that it
would lose the deletion if the only difference for that position is that 80
bases away from it a read is softclipped.
If the deletion was in a clipped position that would be an obvious
explanation, but you said that wasn't the case.

Maybe Strelka takes into account read length and because the matching
region is shorter it thinks it is more likely to be mismapped.

Do you know anything about how Strelka works or determines it's results?
That might help figure it out.

I had asked previously about flag differences in case clipOverlap made a
read unmapped due to orientation clipping. Because that could be a reason
for losing evidence for the deletion. But it sounds like you checked and
that isn't the case.

Let me know if I'm misunderstanding your question.
Fyi there is a bamUtil diff tool (with --all option) that may be helpful
for comparing bams if you need it. Unfortunately it doesn't have an option
to limit the diff to a specific region. I will try to add that in the
future.

—
Reply to this email directly or view it on GitHub
#20 (comment).