per tile sequence quality decreases after QC with sickle

Question

[mary]

I'm trying to QC my illumina miseq paired end reads with sickle pe. but, as visualized in Fastqc per tile sequence quality, it seems that I can not get rid of low quality sequences from special tiles (2118) but somehow it even get worse by increasing quality threshold.

this is the screen-shot from my reads per tile sequence quality after trimming with default quality threshold of sickle pe (q 20):

https://postimg.org/image/s8igtzzc5/

and this is what I get after increasing that to q=30:

https://postimg.org/image/q30kiflbr/

any Idea about what happens around 2118?

[GenoMax]

Instead of dropbox links consider posting the images on imgur.com (or other free image hosting sites) and including the links in your post (use the ctrl+G option when editing the post).

Answer 1

[Brian Bushnell]

I suspect it turns red if 100% of the reads are gone from a tile, in which case the calculated average quality would be 0 to avoid a division by zero problem.

For this kind of issue, I suggest using FilterByTile rather than blanket-applying extremely high quality-trimming thresholds like Q30 to your entire dataset, which incurs bias and can greatly damage the utility of the data for many purposes.

[mary]

seems reasonable! thanks for your help!