GATK recalibration problems: low coverage and few SNVs

We've had reports about very strange results mostly on viral genomes. In such cases very few SNVs will be reported, with odd frequencies and very low coverage. We could trace this back to GATK's base call recalibration step, in which for yet unknown reasons, most base call qualities are set to the lowest possible value by GATK. We are investigating this and ask users to run LoFreq on the uncalibrated data in such cases (keeping in mind that spurious SNV calls are possible).

[Andreas Wilm]

This seems to happen when you're using a fasta reference sequence, which contains white-spaces. This previously popular format separates blocks of ten or so nucleotides using a space. GATK (at least 2.0.X) happily works away on it, but produces garbage results (no error message; base call qualities set to a minimum) and all downstream programs therefore produce unusable results as well. Simply remove any spaces in your reference sequence before running GATK.