This site is now read-only. You can find our new documentation site and support forum for posting questions here.
Be sure to read our welcome blog!
Base quality score recalibration makes all reads low quality?
TomJKono
Member ✭
Hello,
I am running the GATK pipeline on barley exome capture data. In order to verify that we are doing a good job with realigning around indels and are calling true variants, we are comparing our alignments to known variants from Sanger resequencing. We are using the latest version of consed (26) to view our BAM files. However, we noticed that after base score quality recalibration, almost every base seems to be marked as Phred20. Before recalibration, our scores were Phred30 and above, and our alignments looked relatively clean.
Is there an option in the base quality recalibrator tool that would affect this? How can we avoid tamping down our quality scores so aggressively? Thank you.
Best Answer
-
Geraldine_VdAuwera
Cambridge, MA admin
Yes, possibly -- it's definitely worth a shot to try and bootstrap a set of variants from your data in addition to the knowns you have.
Answers
It's hard to say, it depends on the data. What did the pre/post recalibration plots look like?
Thank you for the fast response. Attached are the plots generated from AnalyzeCovariates and our recalibration report. It looks like we have reads with reported qualities up to Phred40, but for some reason, they are being adjusted down to about Phred25.
Based on the plots there are two possibilities: either the base quality calls in your data are overestimated (the machine was too optimistic about its own accuracy) and therefore the recalibrator is correct to tamp down the qualities, or the set of known variants you are using is insufficient and a lot of true variation is getting counted as errors. What are you using as known variants?
We only have a set of about 8,000-9,000 known variants from a SNP genotyping array, and a handful of Sanger resequenced loci. Could this be a problem? Would calling SNPs, and using the high confidence variants in a second round of the BQSR help?
Yes, possibly -- it's definitely worth a shot to try and bootstrap a set of variants from your data in addition to the knowns you have.
Thank you, we will try that and report back here.