[mira_talk] Re: Homopolymer errors and MIRA
- From: Lionel Guy <guy.lionel@xxxxxxxxx>
- To: mira_talk@xxxxxxxxxxxxx
- Date: Wed, 30 Sep 2009 17:19:49 +0200
Hi Bastien,
On 26 Sep 2009, at 17:20 , Bastien Chevreux wrote:
hrm, let me guess: Newbler 2.x?
Yes, Newbler 2.0.0.20
What I'd be interested in would be this: do you have some statistics
which
show the errors broken down by homopolymer length. I strongly
suspect that
longer homopolymers are more prone to the base calling error, so
shifting
calling weights according to homopolymer length is probably one
possible
solution.
It doesn't seem so. Actually, it seems that the absolute number of
overcalling errors is constant over hp length, but that the number of
undercalls is going down with hp length (this is only for mira,
numbers of homopolymers with length (rows) and length diff with
reference (columns):
-2 -1 0 1 2
All 4 2 80 62502 7 1
5 2 93 22854 5 2
6 2 156 7819 7 0
7 2 110 2596 6 0
8 1 21 621 7 0
9 0 0 49 4 0
Ideally there would also be statistics which show how many
gaps/bases were at each erroneous site, but that might be a bit too
much to
ask.
I attach a file to this email, which contains (for mira assembly for
one of the genome) each "incorrectly" called HP, and every read as
they appear in the assembly file. Hope that helps... If you want more,
or even the full dataset (including correct calls), or the scripts,
let me know...
Lionel
============================================
Lionel Guy
Thunmansgatan 25, SE-75421 Uppsala
phone: +46 (0)18 245596
mobile: +46 (0)73 9760618
email: guy.lionel@xxxxxxxxx
============================================
Other related posts:
