SEQanswers

Go Back   SEQanswers > Sequencing Technologies/Companies > Ion Torrent



Similar Threads
Thread Thread Starter Forum Replies Last Post
Ion Torrent $1000 Genome!? Benchtop Ion Proton Sequencer aeonsim Ion Torrent 88 10-28-2012 04:50 AM
Ion torrent blog on raw accuracy - good primer for accuracy grand challenge lek2k Ion Torrent 3 09-20-2011 12:13 AM
Ion Torrent Accuracy Problem and Challenge - Non-technical explanation lek2k Ion Torrent 0 09-13-2011 09:12 PM
Challenges in Improving Ion Torrent Raw Accuracy lek2k Ion Torrent 0 08-21-2011 11:31 PM
ion torrent herrroaa Introductions 5 07-25-2011 05:36 AM

Reply
 
Thread Tools
Old 08-28-2011, 01:54 AM   #1
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Default Ion Torrent Rapid Accuracy Improvements (independent analysis)

The new blog series covers improvements over a 3 month time frame using the various data releases and includes in house PGM runs.

http://biolektures.wordpress.com/201...-improvements/
lek2k is offline   Reply With Quote
Old 08-29-2011, 03:25 AM   #2
pmiguel
Senior Member
 
Location: Purdue University, West Lafayette, Indiana

Join Date: Aug 2008
Posts: 2,309
Default

Very interesting. About the quality value discrepancy: for your empirical QVs, were you using a gapped alignment method? If so, I wonder if that is the source of the marked difference between projected QVs and actual accuracy?

--
Phillip
pmiguel is offline   Reply With Quote
Old 08-29-2011, 06:25 PM   #3
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Smile

Yes, a gapped alignment was used. Using the following command line options
-g 20 -x5

which corresponds to a gap opening penalty of 20 and extension penalty of 5.
The default is -g 40, -x 15.
http://bit.ly/pfxhgj

I may be wrong but I thought the predicted QV (or projected) is based on a prediction algorithm/method. For Ion Torrent, it reads in a phredTable of precalculated values for a range of QVs. I'll blog about this one day.

Therefore, the predicted QV is independent on what goes on with the empirical QVs based on alignment. FYI: Illumina MiSeq is pretty much spot on with their predicted vs actual. I think that's because the technology it's built on has been around for a while thus their predictive algorithms have had time to mature.... Some one correct me if I am wrong, could be just full of crap
lek2k is offline   Reply With Quote
Old 09-02-2011, 09:45 AM   #4
pmiguel
Senior Member
 
Location: Purdue University, West Lafayette, Indiana

Join Date: Aug 2008
Posts: 2,309
Default

Yes that sound right to me.

What I meant was that if you are calibrating your quality values based on agreement between a set of your reads and a known sequence, then you would need to do an alignment to that reference.

Then you check each base in a given QV bin and see if it is was right or wrong. If the proportion of wrong:right is correct, great. If not, adjust your QV table.

Right? But what if you used something other than the default Novoalign settings. Or used another aligner altogether? Without gaps in your alignment, every indel causes every base downstream to disagree with the reference. It would cause you to recalibrate your QV table and drastically reduce the quality of your bases.

You give a link to Novoalign -- is that the official alignment engine of the Ion Torrent, or is it possible they are using a more gap-draconian alignment methodology?

I don't have any particular expertise in calibrating quality values. But it just occurs to me that indels may be the cause in the descrepancies between the QVs assigned and the actual accuracy of the reads with your aligner.

What do you think?

--
Phillip
pmiguel is offline   Reply With Quote
Old 09-02-2011, 01:29 PM   #5
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Default

Novoalign is not the official alignment program. tmap is the official one created by Nils Homer so he would be the best to answer any questions on that.

Regarding QV calibration, I'm quite new to the concept so would need to brush up before I comment. I will look at the QV prediction code then ask on the Ion Community to make sure I've understood it before blogging what I've learned.
lek2k is offline   Reply With Quote
Old 09-04-2011, 09:27 AM   #6
pmiguel
Senior Member
 
Location: Purdue University, West Lafayette, Indiana

Join Date: Aug 2008
Posts: 2,309
Default

I will be interested to read what you find.

Seems like there is a real issue here caused by the high indel error frequencies of 454 and Ion Torrent. To a first approximation it is completely legitimate to allow indel miscalls resulting from incorrect estimations of the numbers of bases in a run to produce much lower quality values. But because utilizing gapped alignment procedures can side-step most of the issues caused by indels of this sort, one doesn't want to penalize the QVs too much.

--
Phillip
pmiguel is offline   Reply With Quote
Old 09-04-2011, 10:19 PM   #7
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Default

Broadly, this is how QVs are predicted using six metrics and a phred lookup table. This is detailed in my latest blog post.

http://biolektures.wordpress.com/201...ion-algorithm/
lek2k is offline   Reply With Quote
Old 09-06-2011, 04:49 AM   #8
pmiguel
Senior Member
 
Location: Purdue University, West Lafayette, Indiana

Join Date: Aug 2008
Posts: 2,309
Default

Hi lek2k,
Would it be possible to generate the plots in the figures using a gapless alignment method? Just for comparison?
One possibility here is that Novoalign is calling correct bases that Ion Torrent would score as miscalls. Every base call after an indel is a miscall if you do not use a gapped aligner.

--
Phillip
pmiguel is offline   Reply With Quote
Old 09-13-2011, 09:15 PM   #9
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Default

Great EdgeBio blog post on Predicted vs Empirical QV and presents recalibration using GATK

http://www.edgebio.com/blog/?p=271
lek2k is offline   Reply With Quote
Old 09-20-2011, 12:11 AM   #10
lek2k
Member
 
Location: Australia

Join Date: Aug 2011
Posts: 32
Default

Quote:
Originally Posted by pmiguel View Post
Hi lek2k,
Would it be possible to generate the plots in the figures using a gapless alignment method? Just for comparison?
One possibility here is that Novoalign is calling correct bases that Ion Torrent would score as miscalls. Every base call after an indel is a miscall if you do not use a gapped aligner.

--
Phillip
Yes that would be interesting for comparison. I'll give it a go soon.
lek2k is offline   Reply With Quote
Reply

Thread Tools

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off




All times are GMT -8. The time now is 04:16 PM.


Powered by vBulletin® Version 3.8.9
Copyright ©2000 - 2019, vBulletin Solutions, Inc.
Single Sign On provided by vBSSO