De novo Assembly k-mer question

[ronaldrcutler]

Hello all, I have a general discussion question:

If small kmers are used in an assembly that utilizes a de bruijn graph, we can expect to get many more contigs than we would if we used larger kmers. However, with smaller kmers, the contig quality is usually higher, sorting out repetitie elements, etc.

In the assembly pipeline, when scaffolding, we are using the paired-end reads to essentially put together fragments that may not have had enough coverage or overlap to join in the assembly. My question is: if we use small kmers in the initial assembly and then scaffold, will this result in the same amount of scaffolds if we were to use larger kmers in the initial assembly and then scaffold? What is the effect on the contig quality - if any?

Cheers

[Brian Bushnell]

Small kmers do not necessarily give a higher-quality assembly, with a normal assembler (ignoring multi-k assemblers for the moment). Rather, longer kmers should increase both continuity and quality, for a dataset with fairly uniform coverage. I recommend using whatever value of K gives you the greatest continuity in your initial assembly.