dubious SNP calls from GATK

[caswater]

Hi, I followed the Best Practice v3 steps and called SNPs with GATK. After having the raw SNPs, I used VQSR to recalibrate and post-filter the reported raw SNPs. However, the trenches plot from VQSR recalibration looks like below:



which indicates that even if I used the recommended cutoff to refine the SNPs, I still get a lot false positives.

My data are just standard Illumina data, my analysis just followed GATK Best Practice v3. Do you have this problem before, and how shall I solve this issue?

Thanks a lot