Honestly Sebastien, comparing Velvet to Ray hand to hand. Velvet has a
nucleotide coverage to optimal kmer. So if i know my genome is well
covered, I can guess a good kmer for velvet. For Ray, what makes a good
assembly?
trial and error, but more than that, what are we basing this trial and
error?
Adrian
Subject: Re: [Velvet-users] Fwd: Re: Assembling NGS data
Date: Thu, 14 Feb 2013 23:33:19 -0500
From: Adrian Pelin <[email protected]>
To: Sébastien Boisvert <[email protected]>
CC: [email protected] <[email protected]>
A fellow Canadian that visits Laval and Montreal on a regular basis
indeed :)
What puzzled me is kmer optimization in velvet. At one point, i had a
good covergae with 250 bp read and you told me to use a smaller kmer
than 240s. If you coverage is awesome, velvet likes kmers close to read
length, that's what Velvet Optimizer suggests, I am wondering what a Ray
optimizer would say. I would need some spare time to understand Ray's
kmer hotspot. (what conditions make Ray optimal, what coverage? what
kmer based on read length?)
But that is besides the point. In my opinion, the people from GAGE did
not take took much time to optimize velvet (using -auto is what you do
to get an idea, NOT as a final results, in my opinion), and the results
they provided for velvet are not reproducible to me (I have gotten
better, tweaking some things).
Thanks for explaining Ray more. An assembler, a robot... if you spent
money on sequencing, your better expect less than magic from assemblers.
DBG, expects at least understanding that different kmers produce
different results.
On 2/14/2013 10:52 PM, Sébastien Boisvert wrote:
On 02/14/2013 10:13 PM, Adrian Pelin wrote:
I am suggesting, but at the same time, if you think I am wrong, please
tell me. I am in Ottawa by the way, so not too far from you.
A fellow Canadian, then ?
I want your input in what GAGE did, because they are showing that velvet
is a shitty a assembler in their tables. And that is not true. Ray has
an algorithm similar to velvet, DBG,
No, Velvet and Ray have totally different algorithms !
The de Bruijn graph is a data structure, not an algorithm. Velvet
simplifies the graph,
Ray does not simplify the graph and instead perform distributed graph
traversals in
parallel.
and my first look at their
"recipes" is that they do not understand velvet....
You need to optimize velvet, you can't compare it with others using
-auto options....
An assembler is essentially a software robot. You
want it to be autonomous -- you give it data, some compute resources,
and it should
be OK on its own for a while.
Yes, indeed they picked a shitty dataset with bad quality "IN MY
OPINION".
The information you provided indicates that the dataset was of bad
quality, and
therefore this is a fact.
But more than that, when I asked them, why did velvet perform
so badly? They said, ask the velvet people... LOL.
hehe
Do you think that the performance was due to -auto ?
On 2/14/2013 9:13 PM, Sébastien Boisvert wrote:
So basically your are suggesting that GAGE gauged error correction
more than assembly.
On 02/14/2013 05:19 PM, Adrian Pelin wrote:
To me, poor sequencing is a Q30 of lower than 70%. I think this means
that less than 70% of reads have an average Q30 (phred score).
This dataset has an Q30 of 14% and 32% for its shortjump and longjump
for the/S. aures/ assembly specifically, so unless I am calculating
these in the wrong way, this is absolutely terrible sequencing.
On 2/14/2013 3:21 PM, Laurent MANCHON wrote:
-------- Message original --------
Sujet: Re: [Velvet-users] Assembling NGS data
Date : Thu, 14 Feb 2013 14:57:12 -0500
De : Sébastien Boisvert <[email protected]>
Pour : [email protected] <[email protected]>
On 02/14/2013 02:46 PM, Laurent MANCHON wrote:
Le 14/02/2013 17:47, Sébastien Boisvert a écrit :
On 02/14/2013 08:53 AM, Adrian Pelin wrote:
Interesting paper here comparing velvet to other assemblers:
http://www.ncbi.nlm.nih.gov/pubmed/22147368
What I do not understand is why velvet performs so poorly
compared to
other assemblers. It seems the authors didn't take the time to
optimize
the settings and use "auto". Another thing I noticed is the poor
quality
of the sequencing.
What in particular is indicative of poor sequence quality in this
case ?
Any other thoughts?
Mira assembler is missing, i don't know why.
I use Velvet and Mira and both are good assemblers, depending on
what
input sequences you work
and what you have to obtain.
Ray is also missing.
p.s. you forgot to CC. the list.
Laurent --
Adrian
_______________________________________________
Velvet-users mailing list
[email protected]
http://listserver.ebi.ac.uk/mailman/listinfo/velvet-users
_______________________________________________
Velvet-users mailing list
[email protected]
http://listserver.ebi.ac.uk/mailman/listinfo/velvet-users
_______________________________________________
Velvet-users mailing list
[email protected]
http://listserver.ebi.ac.uk/mailman/listinfo/velvet-users
------------------------------------------------------------------------------
Free Next-Gen Firewall Hardware Offer
Buy your Sophos next-gen firewall before the end March 2013
and get the hardware for free! Learn more.
http://p.sf.net/sfu/sophos-d2d-feb
_______________________________________________
Denovoassembler-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/denovoassembler-users
