I am working with joint-called WGS samples with mixed coverage (10x to >50x). Variants already passed GATK VQSR.

Cumulative sample groups:

• S1: coverage 20x or higher (N=100)
• S2: coverage 15x or higher (N=120; includes S1)
• S3: coverage above 12x (N=180; includes S1+S2)
• S4: coverage 10x or higher (N=300; includes S1+S2+S3)

For each group:

bcftools filter -S . \
  -e 'FMT/GQ<15 || (TYPE="snp" && FMT/DP<6) || (TYPE!="snp" && FMT/DP<8)'

Then:

bcftools +fill-tags -- -t AC,AN,AF,F_MISSING,NS
bcftools view -i 'AC>0'

Unexpectedly, retained variant counts decrease:

• S1: ~10M
• S2: ~9M
• S3: ~8M
• S4: ~4M

Since groups are cumulative, I expected counts to remain stable or increase.

Can genotype masking in lower-coverage cumulative cohorts realistically reduce AC>0 variants this much, or does this suggest a workflow/filtering issue?