[mira_talk] Re: MIRA v. 3.9.15 vs MIRA v. 4.0.2 performance degradation issues when using Illumina 54-bp paired-end reads in hybrid assembly
- From: Bastien Chevreux <bach@xxxxxxxxxxxx>
- To: mira_talk@xxxxxxxxxxxxx
- Date: Thu, 22 May 2014 22:21:33 +0200
On 22 May 2014, at 21:41 , Bayles, Darrell <Darrell.Bayles@xxxxxxxxxxxx> wrote:
> The bacterium does have a large number of large repetitive elements, and yes
> most are transposons. We have considered PacBio and that would be the
> simplest way to work through the big repeats; however, I’d still like to get
> some clarification regarding the questions of performance differences between
> MIRA v. 3.9.15 and MIRA v. 4.0.2, and clarification about the questions
> resulting from your comments about short reads. While I didn’t expect there
> to be a big improvement in assembling with v. 4.0.2, I certainly didn’t
> expect a substantial decline in the goodness of assembly either.
There are a couple of things to consider when trying to explain the differences
you see. Remember, I do not know the data set, I’m just guessing.
1. No guess here: 3.9.15 belongs to those development version where more
misassemblies happened than in later versions as people had given me a lot of
tough data to optimise MIRA for.
2. the “default” settings have changed all along the 3.9.x development as I
adjusted MIRA to more current data sets. And this means that some heuristics
now are (much) better adapted to seeing “short” reads in the 100+ bp range and
will utterly fail for smaller reads. The “why” I did not bother to investigate,
100bp Illuminas have been here since 2010 or so and why should I spend time
optimising for data sets no one is generating since 3+ years or so?
Especially the failing heuristics probably leads to the very long assembly
times you’re seeing: it’s spending way more time in Smith-Waterman alignments
than I’d expect. Which means that too many hits from the SKIM phase are not
pruned out, which in turn leads to sub-optimal overlap graphs and that leads to
… well, in the end, worse assemblies with shorter reads when you’re using the
4.x series of MIRA.
Maybe I should add another “Nag and Warn” flag which stops the assembly if it
detects a readgroup with reads distinctly smaller than, say, 80bp.
B.
--
You have received this mail because you are subscribed to the mira_talk mailing
list. For information on how to subscribe or unsubscribe, please visit
http://www.chevreux.org/mira_mailinglists.html
Other related posts:
