parameter 'seed kmer' of IDBA_ud

[GUZUMO]

I just don't know what's the meaning of it.
Is there anyone explain it for me ?
Thanks !XD

[Brian Bushnell]

I am not sure about IDBA, but Tadpole has a seed kmer flag, and the meaning is:

Contigs are built starting with a single kmer, then extending it. However, only kmers with a sufficiently high count (number of times they are observed in reads) are used as seeds, because kmers with a very low count are likely to be erroneous. So Tadpole's default "mincountseed=3" ensures that no contig will ever be created that does not have at least one kmer in it with a depth of at least 3. I suspect IDBA uses a similar mechanism.

[GUZUMO]

I am not sure about IDBA, but Tadpole has a seed kmer flag, and the meaning is:

Contigs are built starting with a single kmer, then extending it. However, only kmers with a sufficiently high count (number of times they are observed in reads) are used as seeds, because kmers with a very low count are likely to be erroneous. So Tadpole's default "mincountseed=3" ensures that no contig will ever be created that does not have at least one kmer in it with a depth of at least 3. I suspect IDBA uses a similar mechanism.

Thank you .
But as idba_ud has a parameter '--min count' with description 'minimum multiplicity for filtering k-mer when building the graph', I guess maybe what you say is something like this.[?]
Sorry that I forgot to write down the description of seed kmer, 'seed kmer size for alignment'。 But I really can't understand. Is it means when constructing the graph, the first kmer needs to be this size[seed kmer] and other kmers' length will be the sum of mink and step?

[Brian Bushnell]

Oh... in that case, speculation again, but it sounds like it is for a scaffolding phase; essentially, a match of at least that long is required in a read to consider it as having come from that scaffold.