Manual Quality Control for Variant Calls

[onconaut]

Hello all SEQers,

I am trying to find some documentation or leads on strategies for the manual quality control of variants that have been called in a whole exome sequencing experiment.

The closest I got was in the GATK 'best practices' summary which basically tells you to check if the variants 'look good'. Obviously this leaves a lot of room for interpretation, but this is the closest thing I have found to a documentation of a strategy. All sequencing papers I have found will describe the pipeline and filters to some degree, but then do not touch the topic of manual review and why they would exclude some of the variants.

Can someone in this community point me towards a publication or share their 'secret sauce' in terms of manual review strategies?

Thanks, O.

[HESmith]

'Looks good' is in the eye of the beholder, but common metrics include sequencing depth (more is better), variant present on reads from both strands (strand-specific variants are more likely to be false-positives), mapping quality, and complexity of the alignment (i.e., nearby indels increase the likelihood of false-positive SNP calls).

Blue Collar Bioinformatics blog (here) has benchmarked several variant-calling workflows (including GATK best practices) against the NIST standard Genome-in-a-Bottle. Data and workflows can be downloaded/replicated, so you can see for yourself what constitutes 'good' and 'bad'-looking variant calls.