Disconcordance between chip run for same sample

Hi all,

I am evaluating the panel performance, which was designed based on amplicon enrichment.
However, I found that 2 out of 2000 variants showed disconcordance between the two run.

For example, rs9272346 is AG in one run, while is GG in the second run.
I take a close look at the bam file and found that it is due to slight change of the alt allele, from 7% to 6%.

I tried both UnifiedGenotyper and HaplotypeCaller, both showed disconcordant results.
Command are listed as follows:

java -jar $jatk_bin -T UnifiedGenotyper -R $ref -D extract_snps.vcf --genotyping_mode GENOTYPE_GIVEN_ALLELES --output_mode EMIT_ALL_SITES -A StrandBiasBySample -A AlleleBalanceBySample -ip 50 -stand_call_conf 10 --alleles extract_snps.vcf -L extract_snps.vcf -dt NONE -I GWI201_0609.rs9272346.bam -o GWI201_0609.rs9272346.vcf

java -jar $jatk_bin -T UnifiedGenotyper -R $ref -D extract_snps.vcf --genotyping_mode GENOTYPE_GIVEN_ALLELES --output_mode EMIT_ALL_SITES -A StrandBiasBySample -A AlleleBalanceBySample -ip 50 -stand_call_conf 10 --alleles extract_snps.vcf -L extract_snps.vcf  -I GWI201_0609.rs9272346.bam -o GWI201_0609.rs9272346.vcf

java -jar $jatk_bin -T HaplotypeCaller -R $ref -D extract_snps.vcf --genotyping_mode GENOTYPE_GIVEN_ALLELES --output_mode EMIT_ALL_SITES -A StrandBiasBySample -A AlleleBalanceBySample -ip 50 -stand_call_conf 10 --alleles extract_snps.vcf -L extract_snps.vcf -I GWI201_0609.rs9272346.bam -o GWI201_0609.rs9272346.vcf

Only UG and without downsampling can make the same calling between the two bam file.

We calculate the AB ratio = #(alt)/#(all), and found that
hom ref: 0.0011768167108-0.0630291627469
het: 0.0738045738046-0.691937424789
hom alt: 0.984477124183-1.0

It is obvious that little gap is existed between hom_ref and het variants, which might result in the different genotype calling in our situation.

I am wondering how can I fix this?

Thanks.