How exactly is the _best_ kmer size selected by (meta) spades?

[sivico26]

Hello developers,

I am developing a workflow that uses meta spades as a first assembly step. I have noticed that for essentially all the datasets I have run, the biggest kmer size is always selected to proceed with the following stages.

I find this a bit puzzling, so I wanted to ask how exactly meta spades scores and ranks the assemblies it gets from each kmer assembly? My understanding is that the most contiguous is selected, but how is this determined exactly? Is it the number of nodes or edges in the assembly graph? longest mean size of any of those?

Additionally, in your opinion, what would be the simplest way to infer the k-mer size that was selected (in order to automate the inference)? I do not think parsing the log would be the simplest approach. I am currently looking at which folder contains the final_contigs.paths file, something akin to spades_folder.glob("K*/final_contigs.paths"). Can you think of a better way?

Thank you for your time.

Kind regards,
Sivico

[asl]

I find this a bit puzzling, so I wanted to ask how exactly meta spades scores and ranks the assemblies it gets from each kmer assembly?

Short answer (maybe even more puzzling): it doesn't

Long answer: SPAdes uses all information from all k-mers lengths for a final assembly. So, you should not guess, parse logs or invent some other incorrect ways of determining. Please refer to SPAdes manual that describes the SPAdes output: https://ablab.github.io/spades/output.html

In particular the following is useful:

K<##>/ - directory containing intermediate files from the run with K=<##>. These files should not be used as assembly results; use resulting contigs/scaffolds in files mentioned above.

Hope this helps.