I think I had a misunderstanding about the nature of mean read accuracy. As currently outlined in the blog post by Walter De Coster these are calculated not as the arithmetic mean. Instead the correct way is to first convert Phred scores to probabilities, calculate average error probability and convert average back to Phred scale (as implemented in nanoq).
However this is not the same issue that is raised in the technical basecall accuracy section of the ONT technical documentation (screenshot because walled behind community access)
My reading of this is that the non-arithmetric mean read accuracy should also be applicable to any other sequence with quality scores, and is not specific to ONT reads only; this would mean nanoq can be used for Illumina reads as well.
I think I had a misunderstanding about the nature of mean read accuracy. As currently outlined in the blog post by Walter De Coster these are calculated not as the arithmetic mean. Instead the correct way is to first convert Phred scores to probabilities, calculate average error probability and convert average back to Phred scale (as implemented in
nanoq
).However this is not the same issue that is raised in the technical basecall accuracy section of the ONT technical documentation (screenshot because walled behind community access)
My reading of this is that the non-arithmetric mean read accuracy should also be applicable to any other sequence with quality scores, and is not specific to ONT reads only; this would mean
nanoq
can be used for Illumina reads as well.