I think the developer just took the equation from Heng Li's MAQ faq at http://maq.sourceforge.net/fastq.shtml

$Q = 10 * log(1 + 10 ** (ord($sq) - 64) / 10.0)) / log(10);

which you notice is missing a forward parenthesis somewhere, and which I think was originally for Solexa+64->Sanger conversions anyway. See FASTQ wiki

you see on line 1533 of prinseq-lite.pl:

return ((10 * log(1 + 10 ** (ord($qual) - 64) / 10.0)) / log(10));

he just tacked on an extra ( before the first 10, which means an Illumina phred-64 "B" returns a 10 and a "h" returns a whopping 390

I suspect the correct equation might be:

$Q = 10 *  log(1 + 10 ** ((ord($qual) - 64) /  10.0)) / log(10);

the sad thing is that incorrect equation has been copy-pasted all over the web and even in textbooks

at any rate

return (ord($qual)-64);

should suffice

Is this Illumina 1.3 (phred+64) data? From your mention of B runs I suspect it is:

-si13

Quality data in FASTQ file is in Solexa/Illumina 1.3+ format and should be scaled to Phred quality scores ranging from 0 to 40. (Not required for Solexa/Illumina 1.5+, Sanger, Roche/454, Ion Torrent, PacBio data.)

I haven't used this program, but intuitively, I suspect the [?]trim_qual_left[?] and [?]trim_qual_right[?] work by clipping bases from each side until a base with quality 5 or higher is encountered. So you would end up with reads that might still contain low quality bases, but the two bases remaining at the edges of your reads would have a quality of 5 or higher?

Are you trying to remove all reads where any base has a quality of <5, even if the base is in the middle of the read and is flanked by high quality bases? This seems like an unusual thing to want to do. Normally you would want to limit yourself to trimming the ends since that's where you would expect the low quality bases to be (e.g. in an Illumina experiment, you might see the 5' ends of your reads have high quality, and the quality degrades as you move closer to the 3' end). Read aligners tend to be tolerant of one or two low quality bases in the middle of an otherwise good sequence (configured by parameters).

If you want to remove any read with a base with a quality < 5, you might want to try experimenting with the other trimming parameters. E.g. I might start with setting the [?]trim_qual_window[?] to the length of your reads. From what I understand after a quick look at the manual, this might implement the trimming rule by considering the whole read at once instead of a sliding window of 1 base.