Weird results at the 51st nucleotide

[sulicon]

Hi all,

We just got some bisulfite sequencing data from a set of patients. After running FastQC on a few samples, we observed an extremely low quality score at position #51, for two samples.

Attached are the distribution of base quality scores, % of Ns, and the nucleotide composition at each position.

My question is, what is the possible reason for this? Should we trim all the nucleotides after 50?

Thanks in advance!

[dpryan]

There was probably a bubble that floated through the flowcell. That's a pretty common cause of random dips in quality. I wouldn't bother trimming that off, the dip in quality is only affecting a few bases. Any of the common aligners should still be able to accurately align the reads. Just choose one that allows you to set a minimum phred score during methylation extraction (i.e., one that will ignore methylation calls at those crappy bases).

[sulicon]

Thank you Devon for the explanation and suggestion.

[GenoMax]

Were these two samples on the same lane? Were there other samples besides these two in the lane that are not showing this problem?

You should at least request your sequence provider to re-run the sample(s) at no cost (unless there were other samples in that lane that do not have this problem).

[sulicon]

We got the data from another group. There is no barcodes used, I assume each sample was sequenced on one entire lane. (right?)

We have checked about 10 samples, two samples mentioned above have the problem at exactly the same position (#51). The two samples are in the control group.

There are also other samples with problem at different position. Three of the case samples we checked have low quality base-calling at position #63 (see attached).

It seems that the position of abnormal base-calling differs in a run-by-run manner.

Were these two samples on the same lane? Were there other samples besides these two in the lane that are not showing this problem?

You should at least request your sequence provider to re-run the sample(s) at no cost (unless there were other samples in that lane that do not have this problem).

[GenoMax]

You are correct in that your samples must have been run in separate lanes, if there are no barcodes. It is possible that one or more lanes had a bubble flow though that affected the basecalls. You can look at the tile representation from FastQC to see if you can identify bad tiles across cycles.

In general Q-scores take a nose dive when the nucleotide diversity no longer exists (or is significantly reduced). It appears that majority of the data in the example above consists of N's beyond cycle #63 and would likely need to be trimmed before analysis.

[sulicon]

Thank you GenoMax. I will trim the low quality bases as you suggested.