PrintReads introduces N bases when encoding some CRAMs and changes sequence about gatk

[gokalpcelik]

These reads don't even look the same ?

GTGGTTTTTTTNTNTTTTGTTTTTTTNTTTTTGTGTTTTGTTTTTGTGTTTGTT
TCTTCACCAAAGAGCCCCTAAAACCCGCCACATCTACCATCACCCTCTACATCA

[ilyasoifer]

That is true! Should I update the title of the issue to reflect this?

[gokalpcelik]

Can you check the REF_CACHE? Or alternatively you may provide samtools the reference fasta during cram view.

[ilyasoifer]

Thanks @gokalpcelik!
Seems that -T does not affect much:

samtools view -T /data2/reference/Homo_sapiens_assembly19_1000genomes_decoy/Homo_sapiens_assembly19_1000genomes_decoy.fasta ultMerge.mt.gatk_printreads.cram | grep "038958_1-Z0011-5346565226"
038958_1-Z0011-5346565226	0	MT	3470	60	54M	*	0	0	GTGGTTTTTTTNTNTTTTGTTTTTTTNTTTTTGTGTTTTGTTTTTGTGTTTGTT	DDDDDDDDDDDDDDDDDDD:DD:DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD	AS:i:54	X1:i:0	XD:Z:CATGAGG_GGTGATC	a3:i:92	bi:Z:5346565226f1:Z:CATGAGG	f2:Z:GGTGATC	i1:Z:Q44	i2:Z:Q27	pr:i:22	pt:i:15	px:i:3813	py:i:1262	rq:f:0.03	si:i:3750	tm:Z:AQ	tq:i:195	MD:Z:0T0C0T0T0C0A0C0C0A0A0A0G0A0G0C0C0C0C0T0A0A0A0A0C0C0C0G0C0C0A0C0A0T0C0T0A0C0C0A0T0C0A0C0C0C0T0C0T0A0C0A0T0C0A0	NM:i:54	RG:Z:Z0011

[cmnbroad]

@ilyasoifer Is there any way I can access the original cram (or better yet, a small subset thereof consisting of just MT) that illustrates this issue) and the reference ? It might be hard to debug without that. If thats not possible, a few suggestions: can you try using PrintReads to write the original cram (I would try just MT) first to a cram, then to a sam, and also the original cram to a sam, and see how those compare? It would also be useful to see what that read looks like if you use samtools view on the ORIGINAL cram. Do you know what software/version was used to write the original cram ?

[ilyasoifer]

Hi @cmnbroad - thanks! I can upload to one of the buckets that are shared between Ultima and the Broad.
Could you provide your gcp account so I can give you permissions?
I am guessing that Megan Shand can share it through our joint slack channel if you prefer.

The original file was created using samtools v1.17.
I provided above the result of writing BAM and Cram with PrintReads. Is it helpful or you prefer SAM? And if I just do
samtools view ultMerge.mt.cram I get

038958_1-Z0011-5346565226	0	MT	3470	60	54M	*	0	0	TCTTCACCAAAGAGCCCCTAAAACCCGCCACATCTACCATCACCCTCTACATCA	DDDDDDDDDDDDDDDDDDD:DD:DDDDDDDDDDDDDDDDDDDDDDDDDDDDDDD	bi:Z:5346565226	rq:f:0.03	pr:i:22	pt:i:15	px:i:3813	py:i:1262	si:i:3750	tq:i:195	tm:Z:AQ	i1:Z:Q44	f1:Z:CATGAGG	i2:Z:Q27	f2:Z:GGTGATC	a3:i:92	XD:Z:CATGAGG_GGTGATC	X1:i:0	AS:i:54	MD:Z:54	NM:i:0	RG:Z:Z0011

[cmnbroad]

@ilyasoifer [email protected]. And don't worry about doing the PrintReads conversions I requested - if I have access to the original file and the reference I can debug this directly.

[ilyasoifer]

@cmnbroad, OK, great. Shared the files with Megan, she will transfer them over!

[ilyasoifer]

@cmnbroad - the reference is from here: gs://gcp-public-data--broad-references/Homo_sapiens_assembly19_1000genomes_decoy/

[cmnbroad]

Thanks for reporting this @ilyasoifer - it looks like a serious bug. I see whats going on and am working on a fix.

[ilyasoifer]

Thank you, good to hear that it was useful

[ilyasoifer]

@cmnbroad - hope all is well. I was wondering if there is any ETA when the fix that you made will be released?

Thanks!
Ilya

[droazen]

@ilyasoifer It will be part of the GATK 4.6 release, which should come out this month. We've also implemented a cram scanning tool that users can use to scan their crams to see if any are affected.

[ilyasoifer]

Ah, great, thanks for updating me!

[droazen]

Summary information about this bug:

This issue affects GATK versions 4.3.0.0 through 4.5.0.0, and is fixed in GATK 4.6.0.0. The PR with the fix is: #8900

This issue also affects Picard versions 2.27.3 through 3.1.1, and is fixed in Picard 3.2.0.

This bug is triggered when writing a CRAM file using one of the affected GATK/Picard versions, and both of the following conditions are met:

• At least one read is mapped to the very first base of a reference contig
• The file contains more than one CRAM container (10,000 reads) with reads mapped to that same reference contig

When both of these conditions are met, the resulting CRAM file may have corrupt containers associated with that contig containing reads with an incorrect sequence.

Since many common references such as hg38 have N's at the very beginning of the autosomes and X/Y, many pipelines will not be affected by this bug. However, users of a telomere-to-telomere reference, users doing mitochondrial calling, and users with reads aligned to the alt sequences will want to scan their CRAM files for possible corruption.

The other mitigating circumstance is that when a CRAM is affected, the signal will be overwhelmingly obvious, with the mismatch rate typically jumping from sub-1% to 80-90% for the affected regions, making it likely to be caught by standard QC processes.

A CRAM scanning tool called CRAMIssue8768Detector that can detect whether a particular CRAM file is affected by this bug was added in #8819, and was released as part of GATK 4.6.0.0