SNP indel variant quality metrics and filtering (v2.6 or newer)

v2.6 has all the same metrics as v2.5 and older, see page for v2.5 and older first. In v2.6 we introduced several new quality metrics tailored to multi-allelic variants that are specified for each alternative allele, instead of one metric for all alleles. These are:

• QDalt, ALT quality by depth
• SBalt, read strand bias per ALT
• MMalt, mismatch percentage in the graph alignment per ALT
• CRalt, clipped bases percentage in the graph alignment per ALT
• SDalt, alignment score difference (AS-XS) given by mapper per ALT
• MQalt, root-mean square mapping quality per ALT

We also trained a new logistic regression classifier, INFO/AAScore, which gives a score between 0 and 1 for each alternative allele.

For single sample calling it is recommended to keep only on the "FILTER=PASS" variants and additionally, in larger genotyping runs, we suggest to keep only alternative allelese with INFO/AAScore > 0.5. This threshold might be adjusted for different applications.