Strange FastQC "Per base sequence content report"

[tu.le]

Hi,
I would like to ask if any one experienced with the kind of "Per base sequence content" graph below generated by FastQC? and what does it mean? If only the first ~80 bases are kept does it decrease the coverage and affect downstream analysis?
Thanks in advance,

[gringer]

These wouldn't happen to be paired-end reads separated by a barcode sequence around 80bp would they? The base proportions for complementary bases flip after that point.

Despite that, the zero counts are a little odd -- A GC content of 10% seems a little on the low side.

[tu.le]

Hi Gringer,
As far as I know this is not PE but SE library. Also, the bias toward low GC content may be caused by BS treatment. Whether removal of bases after flipping point decreases the coverage and/or affects downstream analysis?

[gringer]

The flipping is very suspicious. I would certainly treat the halves separately, but first try and split the reads at the halfway point and map both halves. If you find that the second half doesn't map anywhere (including primer sequences), then maybe it will be appropriate to drop those bases.

Finding the cause of that flipping would be great, but that's probably going to be quite a challenge.

[GenoMax]

Have you asked your sequence provider if there were other samples on the flowcell which did or did not show this particular phenomenon? That does look odd.

How about the quality plots? Do they look ok?

[tu.le]

@Gringer: Thanks for your suggestion, I'll try to see whether the second halves can be mapped or not.
@GenoMax: This phenomenon was seen for both case and control samples but I don't really know the reason. The quality plots are quite normal, as below attachment. Thanks.

[biznatch]

I've run about two dozen RNA- and ChIP-seq samples and never seen the quality dip like that in the middle, are you sure that is normal? It's still in the green, it just looks weird to me.

[GenoMax]

There seems to be some issue with this run if both control (non-methylated) and treated samples show this pattern. My hunch is that the problem is not related to your samples. You should ask the sequence provider if there was a technical glitch of some kind during the run. It could very well be a bad kit or a problem with the instrument.

[mastal]

How long ago was this sample run?

Note that FastQC is assuming that the format is Illumina 1.5.

The current version is Illumina 1.9, so FastQC may be rendering the base qualities incorrectly.

The shape of the per base sequence qualities plot looks like what you would expect if R1 and R2 were joined together, as someone has already mentioned above.

[westerman]

How long ago was this sample run?

It can't be that long ago with reads extending out to 170+ bases. Unless ...

... R1 and R2 were joined together, as someone has already mentioned above.

Which is my feeling as well.

[tu.le]

I've tried to map sub-sequences after the flipping point separately but the performance was really bad (~0.6%). It seems that they were miss-created even their quality report was quite nice. Also, this is newly created data thus it should be a good idea to ask the provider. Anyway, thank you all for helpful suggestion.