Describe the issue:
On a specific batch of samples, GQs and QUALs seem to be abnormal.
The GQ and QUAL distributions are bimodal and for variants they are much lower than I would expect. It doesn't seem like there is anything wrong with the calls themselves; I get an expected number of variants. I also can not find anything wrong with the input data. It has high base quality throughout the reads, they are 100bp paired end reads from a NovaSeq with the four value binned base quality scores. This is the visual report for one sample.

Here is an example. I would expect this variant to have a much higher GQ and QUAL. I also have attached deepvariant's channels png for this variant.

chr1    169421916       .       A       G       18.4    PASS    .       GT:GQ:DP:AD:VAF:PL      0/1:17:58:29,29:0.5:18,0,22

Is this expected or is something strange happening here, any insight you can provide would be very appreciated.
Thank you

Setup

• Operating system: Ubuntu 20.04
• DeepVariant version: 1.4 (but also 1.2)
• Installation method (Docker, built from source, etc.): Singularity
• Type of data: (sequencing instrument, reference genome, anything special that is unlike the case studies?)
  Novaseq, 100bp paired, HG38

Steps to reproduce:

singularity run -B /usr/lib/locale/:/usr/lib/locale/ -c --pwd $(pwd) -W $(pwd) -B $(pwd) docker://google/deepvariant:1.4.0 /opt/deepvariant/bin/run_deepvariant   --model_type WES   --ref $REF --reads $CRAM --output_vcf $VCF --output_gvcf $GVCF --intermediate_results_dir ./int_results  --regions $BED