I am looking at GATK Best Practices for Data Processing and Germline Variant calling. I see that the workflow calls for base quality score recalibration using BaseRecalibrator [https://software.broadinstitute.org/gatk/documentation/tooldocs/3.8-0/org_broadinstitute_gatk_tools_walkers_bqsr_BaseRecalibrator.php] which assumes that any base that is a mismatch to the reference and not in the variable site DB are sequencing errors (aka low quality). It seems that if a true variant is rare in the population, and not in the variable loci DB, then it would get an artficially low recalibrated quality score.
In your all's experience, has this lead to false negatives when you ave tested the workflow on positive control samples? If not, can you help me reason why it would not lead to missing the true positives that are not masked during the model calibration phase?