New k-mers overpresentation by adapter trimming

[ty23991]

Hi
I am working on an Illumina Hiseq paired end whole genome data.

The data shows overrepresentation of adapter sequences such as TruSeq Adapter, Index 14 (97% over 40bp) according to the quality reported by the fastqc analysis.

So I performed trimming of the reads by the TruSeq-PE adapter sequences in trimmomatic data and performed fastqc after trimming.

The post-trimming data shows appearance of new k-mers overrepresented at various positions within 0-50 bp. The bbmap analysis for the percentage match still shows no remarkable mismatch problems, indicating that the k-mers could be genomic.

I have attached the before and after trimming k-mer overpresentation plots herewith.

This observation makes me doubt the need of trimming in paired end data.

My question is :
Do we indeed need to perform trimming in such cases ?
Any suggestion on change in the trimming was needed such as using only the overrepresented adapter sequences and but not other adapters from the same library in the analysis.

Thanks

[Brian Bushnell]

Considering the Y-axis is log-scale, it looks like trimming vastly reduced the rate of overrepresented kmers (though I've never been sure exactly how to interpret the graph), which would be a good thing. What kind of trim rate did you get?

Also, note that if you have contamination by synthetic artifacts of set length, that will not affect BBMap's error rate histograms because the fully-artifact reads won't map to your reference.

[ty23991]

Hi Brian
Thanks for the suggestion.

Here is the trimming report:

Trimming was performed by using 1 prefix pairs, 10 forward/reverse sequences

Input Read Pairs: 532073495
Both Surviving: 510614940 (~96 %)
Forward Only Surviving: 13002991 (2.4%)
Reverse Only Surviving: 3487288 (0.7%)
Dropped: 4968275 (0.9%)

[Brian Bushnell]

It's hard to extrapolate from that to derive the number of reads/bases that were trimmed due to adapter sequence, but it looks like trimming was probably a good idea in that case.

[ty23991]

Thanks for the idea. I am thinking to perform the alignment on both with and without trimming and compare the results.