Can you please provide
a) GATK version used

4.1.4.1

b) Exact GATK commands used

gatk GenotypeGVCFs -R ref/11-ref.fa -V 11.gvcf.gz -L "11" -O 11-all.vcf.gz --include-non-variant-sites

gatk GenotypeGVCFs -R ref/11-ref.fa -V 11.gvcf.gz -L "11" -O 11.vcf.gz

Possibly related to the following threads:

https://gatkforums.broadinstitute.org/gatk/discussion/24296/why-are-there-s-in-the-vcf-output-of-genotypegvcfs

https://gatkforums.broadinstitute.org/gatk/discussion/9023/is-it-possible-to-suppress-the-non-ref-tag-on-variant-calls

Hello,

I'm trying to generate a VCF file from a single sample that includes every site in the reference genome for iterative mapping and pseudo-assembly. This requires me to inject variants called back into the reference AND to soft-mask sites where there is not enough information to make a call. This means I need GATK to provide me with read depth and confidence at non-variant sites.

However, I'm running into an issue when using --include-non-variant-sites in the GenotypeGVCFs step: for some sites, the <NON_REF> allele remains after this step.

Specifically, in a VCF file with 122052550 sites (including non-variant sites), 9804 sites still have the <NON_REF> allele. If I run GenotypeGVCFs without --include-non-variant-sites, there are 0 sites which still have <NON_REF> afterwards.

Here are some of the lines that still have <NON_REF> after GenotypeGVCFs with --include-non-variant-sites:

11 3227415 . G A,<NON_REF> . . . GT:AD:PGT:PID:PS .|.:0,0,0:0|1:3227415_G_A:3227415
11 3244678 . T TAC,<NON_REF> . . . GT:AD:PGT:PID:PS .|.:0,0,0:0|1:3244641_C_CTCTT:3244641
11 3256761 . A G,<NON_REF> . . . GT:AD:PGT:PID:PS .|.:0,0,0:0|1:3256761_A_G:3256761
11 3256764 . T C,<NON_REF> . . . GT:AD:PGT:PID:PS .|.:0,0,0:0|1:3256761_A_G:3256761
11 3291976 . TGTGGGG T,<NON_REF> . . . GT:AD ./.:0,0,0
11 3291977 . G *,<NON_REF> . . . GT:AD ./.:0,0,0
11 3291978 . T *,<NON_REF> . . . GT:AD ./.:0,0,0
11 3291979 . G *,<NON_REF> . . . GT:AD ./.:0,0,0
11 3291980 . G *,<NON_REF> . . . GT:AD ./.:0,0,0

And here are the corresponding lines in the GVCF file:

11 3227415 . G A,<NON_REF> 0.01 . MLEAC=0,0;MLEAF=NaN,NaN GT:PGT:PID:PS .|.:0|1:3227415_G_A:3227415
11 3244678 . T TAC,<NON_REF> 0.01 . MLEAC=0,0;MLEAF=NaN,NaN GT:PGT:PID:PS .|.:0|1:3244641_C_CTCTT:3244641
11 3256761 . A G,<NON_REF> 0.01 . MLEAC=0,0;MLEAF=NaN,NaN GT:PGT:PID:PS .|.:0|1:3256761_A_G:3256761
11 3256764 . T C,<NON_REF> 0.01 . MLEAC=0,0;MLEAF=NaN,NaN GT:PGT:PID:PS .|.:0|1:3256761_A_G:3256761
11 3291976 . TGTGGGG T,<NON_REF> 0.01 . MLEAC=0,0;MLEAF=NaN,NaN GT ./.

(The last four sites, 3291977-80, were not called as variants in the GVCF.)

While the <NON_REF> strings do cause problems with downstream tools, they are easy enough to remove. But its disconcerting that this would seemingly result in more variants being called with --include-non-variant-sites than without. So any explanation would be helpful!

Another thing I noticed while investigating this is that the vcf file that results from --include-non-variant-sites has fewer sites than the reference FASTA file that was used to generate the GVCF:

122082543 total sites in FA
122052550 total sites in VCF

I would expect them to have equal numbers of sites. Any insights on that would be useful as well.

Thanks so much!