Discrepancy in Variants called when split by chromosomes and not split by chromosomes

Follow

Ramesh

• November 13, 2024 20:57
• Edited

a) GATK version used: 4.6

I have built a pipeline following GATK best practices workflow to call rare germline variants.

I wanted to I wanted to reduce time and so I split the processing at chromosome level - base recalibration(BQSR) and Haplotype Caller for each chromosome followed by concatenation.

When I compared the final concatenated vcf with the one I had processed as a whole, I was able to see that there was a 0.2 ~ 0.5 % difference in the variants called. 

Also for the common variants there was a difference in the genotype for a few of them. I have pasted a handful below:

CHROM_vcf1	POS_vcf1	REF_vcf1	ALT_vcf1	FILTER_vcf1	GT_vcf1		CHROM_vcf2	POS_vcf2	REF_vcf2	ALT_vcf2	FILTER_vcf2	GT_vcf2
<chr>	<dbl>	<chr>	<chr>	<chr>	<chr>		<chr>	<dbl>	<chr>	<chr>	<chr>	<chr>
chr8	94857071	CTT	CT	PASS	0/1		chr8	94857071	CTT	CT	PASS	0/2
chr8	94857071	CTT	C	PASS	0/1		chr8	94857071	CTT	C	PASS	0/2
chr8	132907228	G	A	PASS	0/1		chr8	132907228	G	A	PASS	1/1
chrX	44334764	G	A	PASS	0/1		chrX	44334764	G	A	PASS	1/1
chrX	153911252	CTTTT	C	PASS	0/2		chrX	153911252	CTTTT	C	PASS	0/1
chrX	153911252	CTTTT	CT	PASS	0/2		chrX	153911252	CTTTT	CT	PASS	0/1

 

 

Is this discrepancy expected?

Does base recalibration and/or Haplotype Caller depend on bases across contigs?

 

Edit: When I did base recalibration without splitting the chromosomes, but split only at haplotype caller stage, the vcf files are a 100% match. So I can see that base recalibration is not specific to contigs and uses data across chromosomes. Is this right?

0

• 

  Gökalp Çelik

  • November 18, 2024 13:26

  Hi Ramesh

  You can run BaseRecalibrator on individual chromosomes and combine all recalibration tables to generate a single one for best recalibration results. Differences in BQSR may result in changes in basecall values therefore variant calls may show differences among each run. 

  I hope this helps. 

  1

  Comment actions Permalink
• 

  Ramesh

  • November 18, 2024 17:09

  Hi Gökalp Çelik,

  Thank you for reaching back. 

  So essentially, I can run 

  1. gatk BaseRecalibrator - for individual chromosomes

  2. combine the recal tables into one

  3. Run ApplyBQSR for individual chromosomes using same recal table?

  I will test the combining of recal table and get back to you.

  Thanks again!

  0

  Comment actions Permalink
• 

  Gökalp Çelik

  • November 18, 2024 20:32

  Hi again.

  Answer is yes. Your results should be equivalent once you have the same recalibration table for all contigs. 

  Regards. 

  1

  Comment actions Permalink
• 

  Ramesh

  • November 18, 2024 22:22

  Thank you so much Gökalp Çelik!

  I used GatherBQSRReports for combining and the final result is exactly the same now.

  0

  Comment actions Permalink

Please sign in to leave a comment.