I have a question regarding the BQSR param when running the UnifiedGenotyper. If I leave away that param, will there be a "default" confusion matrix applied for (single-sample) SNP calling?
If you skip BQSR, the unrecalibrated base qualities will be used without modification.
I think you're confused about the usage of the BQSR parameter. It is meant to be used for recalibration of base qualities, and has no direct effect on variant calling calculations. Although it can be applied at the UG step as a matter of convenience, our recommendation is to apply it in a preliminary step with PrintReads, in order to generate an appropriately recalibrated file for archival purposes.
Either way, our best practice recommendation is to apply base recalibration at some point before calling variants in order to ensure the most accurate results possible. If you skip this step, you may be allowing systematic errors to affect your data.
ok, so then I misunderstood it. I thought this would be used for the confusion matrix. If not, so what does the confusion matrix mentioned in http://gatkforums.broadinstitute.org/discussion/1237 look like? Is it hard coded somewhere (I unfortunately could not figure it out in the code) and taken from the supplementary material of the GATK paper (), is it derived from the actual data?
Oh, I see now. You were actually essentially correct in your interpretation, I misunderstood what you were asking because I'm used to considering BQSR and calling entirely separately. But yes, that part of the model uses the recalibrated base quality scores from BQSR, which is where the confusion matrix comes in.
Ok, and what is the default value then when the confusion matrix is not specified by the user? will there be no confusion matrix used at all?