How Can A Base-Called Position Be "Unknown" But Have A Non-Minimal Score?

Question

[Jeremy Leipzig]

Let's say I extract something like this from a qseq or FASTQ file

TTCAGATGTTCATATGCGGATCGGCGCTGGGCCCACGAGATCTAGCAGAGCTCGT.GGGACCACGACCACCGACCC
a`bbbbbbaabbab`ab^`bVa^^bab^[``bba^`]_Ya^`_`^^_Xa\_KYTYD[PY^Y_^[P[V_BBBBBBBB

So the dot is like an N - it can't call the base. So if I look at the FASTQ scores in integer format I would expect that position to have a minimal score. But in fact its score is 'D' or 4, not great but some other called bases at the tail end are 'B' or 2. What gives?

[Chris Miller]

Which platform is this data from?

[Jeremy Leipzig]

illumina - maybe v1.3

Answer 1

[Istvan Albert]

I recall a situation where one of our students (Gue-Su Chang) tracked down a few undocumented behaviors in the Illumina pipeline. Basically once the basecalling is done there are additional filters applied during post-processing to handle a few odd cases. This might be on of those, the score D refers to the original call, but later that gets overridden by another step. I know that's pretty vague. Long story short: I think the score does not apply here.

[Chris Miller]

I'll get on board with this answer. It's pretty easy to envision a scenario where the actual base call and the quality score get adjusted by separate processes at some point.

Answer 2

[Peter]

See this thread: http://seqanswers.com/forums/showthread.php?t=4721

It seems that Illumina can give scores as high as 15 for an N (which may be a bug), and that in their latest pipeline Q2 ("B") is a special marker used at the end of a read with Q4 ("D") the lowest real score (they no longer use Q0, Q1 or Q3).

Answer 3

[Ketil]

I agree this doesn't make much sense. But note also that with standard phred scores, a quality score of about 3 gives a probability of error of about 50%, i.e. it is more likely that the called base is wrong than that it is right.

In almost all cases, I think you should ignore base calls with very low quality.