Weird Kmer Content in Illumina PE data

[rooksie]

We are analyzing a paired-end Illumina library with 150bp reads and we are seeing weird kmer content results from FastQC.

There are several over-represented heptamers and they appear in the same positions in the R1 and R2 samples.

We used trimmomatic to remove adapter sequences and low quality bases, but this didn't seem to fix the issue. We went ahead with a de novo assembly using Abyss and this produced a highly fragmented assembly with more than 60 million short contigs.

We know something is wrong, but we can't explain it.

Are the results of our assembly because of this kmer problem? And if so, what can we do about it? And can anyone let us know what is causing it?

[Richard Finney]

See this paper :

http://nar.oxfordjournals.org/content/early/2015/01/13/nar.gku1341

They focus on miseq but please check out the links to papers for other studies.
The authors note that the "library preparation method and the choice of primers are the most significant sources of bias and cause distinct error patterns. ".

They conclude ... "the library preparation method together with the choice of primers causes an extensive bias towards certain motifs causing substitutions, insertions and deletions,"

"Every time a molecule fails to elongate properly or advances too fast, the overall signal for the cluster suffers from interference. So as the read length increases, the cluster signal can get weaker due to an accumulation of these events resulting in higher error rates towards the end of the read (16). This explains the gradual increase of errors that we observed in the position and nucleotide-specific distributions in addition to the spikes caused by the motifs. "

[nucacidhunter]

I wonder what was the library prep kit, read pairs number and if the kmer plots are from raw reads or after processing. I assume sequencing was done on HiSeq 3000/4000.