Help with megahit K-mer sizes

[felipehcoutinho]

Hi all, questions for those who use megahit for assemblies: Do you seen significant improvements in assembly quality (with regards to getting better N50 and overall longer contigs/scaffolds) when changing the k-mer sizer and k-mer step parameters? If so, which values do you recommend?

Thx!

[franlat]

The theory says that both using small and large k-mer sizes have its own advantages and disadvantages.

Small k-mers (21, 31, 41)
• Advantages: Allow assembly of low-coverage regions, improve recovery of rare sequences, and help reconstruct highly variable regions. They generate more overlapings, improving the requirements in memory usage (less memory required)
• Disadvantages: Can lead to fragmented assemblies, increase the risk of misassemblies in repetitive regions, and introduce more errors due to spurious/wrong k-mer overlaps.

Large k-mers (55, 77, 99, 127)
• Advantages: Improve assembly continuity, enhance resolution in repetitive regions, and generate longer contigs.
• Disadvantages: Require higher coverage, may fail to recover low-abundance sequences, and are less effective in highly diverse datasets. They also require more memory because more unique sequences are generated.

Since MegaHit allows the use of multiple k-mer sizes, in our group we use a minimum of 27, maximum of 127 and steps of 10. I think this is very standard. However, I had trouble with k-mer 27 in the past in very complex metagenomes, so you can consider increasing the minimum to 37 instead.