Bwa Soft-Clipping Pattern In Read Length Distribution Using "-Q 20" Parameter

Question

[toni]

Hi all,

I just had to analyze a really bad lane. It is an old Illumina GAIIx lane with reads of 152 cycles. The last 30-40 cycles are of really really bad quality on average so I decided to use the -q 20 switch in bwa aln to trim reads 3'ends based on quality prior to mapping (something I usually would not do).

To have a look at what this trimming parameter left in the BAM file I drew the distribution of the length of soft-clipped part in the reads. To do so, I took the CIGAR string for 20 million alignments that were declared unique by BWA (XT:A:U tag). Here follows what I got :

So we can see that there is a periodic pattern after the main pic (representing no soft-clipped bases). We can also notice the slightly higher bar at position 117. 152-117=35, indeed by default BWA won't trim the reads to something less than 35bp.

Have you already noticed such a pattern using bwa aln -q and what in the algorithm produces this ?

Because looking at the -q definition in the doc, I can not see any reason why the trimming would have such a periodic pattern.

Thanks guys.

T.

[Martin A Hansen]

This I have noted before. It comes from the Illumina sequencing: The Meaning Of B In Illumina 1.5 Pipeline Data?

[toni]

Thank you for this information.

[Istvan Albert]

awesome text plots! ;-) could you please add this as an answer as well?

[Istvan Albert]

Is it possible that the qualities in the file have some sort of periodicity?

[toni]

I do not think so, but now that you raised the point, I will probably do a few checks on this :)

[toni]

By looking at the link given in the comment below, it seems that you were right !

Answer 1

[Martin A Hansen]

This has been noted before and Illumina tech support answered this:

The cyclic nature is a result of several factors:

The neighborhood analysis aspect of Illumina Q-scoring results in a 5-cycle cycle nature to the scores. This has been noted in the past. The other aspect is the quality of the sample. In certain cases this cyclic pattern may be more pronounced in some samples. The use of the Trim_Seq option may also result in a heightened presentation of the cycles, especially in cases of more aggressive trimming.

[toni]

I accept this as the right answer, because I don't think we will get a deeper explanation of this "problem". But that is enough to understand the putative origin of this pattern. Thanks again.