Build Graph lost k-mers

Question

[lardo]

I have a FASTA file contains unique k-mers with the read length is the k-mer size:

>kmer_1
ATGACAGCCTTTTTTAAA
>kmer_2
ATGACAGCCTTTTTTAAT

Then I used the API of gatb-core-1.0.6-Linux :

Graph::create ((char const *)"-in %s -kmer-size %d -abundance-min 1 -nb-cores %d -out %s", xxxx);

There should be 11,571,887 unique k-mer in my file, but the graph build by this file contains only 11,065,132 unique k-mers.

I think this program lost some useful k-mers while storing k-mers.

Answer 1

[edrezen]

Hello,

It is possible that you have N characters inside your data. In such a case, no valid kmer can be built, so the read having one or several N won't be used.

Could you check this is the case of your input ?

[lardo]

No 'N' character in the reads.The lost k-mers can be found in my file but not contained in the graph.

Answer 2

[Brian Bushnell]

Perhaps the program is storing reverse-complements in a canonical fashion, so they are only represented once. That's fairly typical.

You can count kmers with BBTools and, using the 'rcomp' flag, enable or disable storing of kmers and their reverse-complements independently, to get the count each way:

kmercountexact.sh in=file.fasta k=18 rcomp=t
kmercountexact.sh in=file.fasta k=18 rcomp=f

[Rayan Chikhi]

Yes GATB does collapse each k-mer and its reverse complement into a single canonical kmer.

[Rayan Chikhi]

If this answer does not solve your problem, would you mind posting the dataset? (I'm assuming this is a ~200 MB file, possibly much less if gzipped)