[Denovoassembler-devel] RE : RE : RE : Why flush both k-mer and its complement?

[Sébastien B. <seb...@ul...>]

flush_twice_nocomplement_v2.patch


This patch has 1 bug: rankToFlush must be computed with the lowest k-mer.

There is 1 reverseComplement in the design I described in the other email.

Basically, you don't call vertexRank, instead, you call the hash function on the lowest directly.



vertexrank_nocopy_nocomplement.patch


The method vertexRank is used elsewhere too, in places where reverseComplement is
not called before this call. 


> 
> Attached. Note that this is actually 2 reverse-complements, not 1,
> because a reverse-complement is needed in order to get the correct
> vertex rank. I suppose you could avoid this by modifying the vertex rank
> function to take a boolean "actually, don't bother with finding the
> lowest k-mer, I've already done that."
> 
> oh... and looking at the vertexRank function, you could eliminate a Kmer
> copy by using an if statement (see attached patch, which does both of
> these things, but doesn't modify the KmerAcademyBuilder.cpp function to
> take advantage of that).
> 
> FWIW, when I run this (both patches applied) on my phiX_5k data, it
> produces different coverage distributions each time. Does this make sense?
> 
> -- David
                  Sébastien Boisvert
                  http://github.com/sebhtml/ray