How to do the recalibration of non-human species without good SNP database

Xin Fan · 05-23-2012, 01:05 PM

Hi all,

I am dealing with the genome data of one bird, there are only a few studies about this bird (thus I do not have good SNP data to do the recalibration). I wonder how can I do the recalibration.

I found a suggestion from GATK website:
http://www.broadinstitute.org/gsa/wik...

They said we can bootstrap a database of known SNPs. However, I still have some questions.

1. When I do the initial round of SNP calling on my unrecalibrated data, Should I use any conditions or not? My colleagues are genotyping other genomes. When they do the recalibration (also without good SNP database), they are only requiring that genotypes containing an SNV had a genotype quality score ≥10. If I need to do the same thing as they did, what threshold should I use? They have high coverage genome (>20X), whereas my bird genome is low coverage (15X, after running DepthOfCoverage).

2. They also said "These steps could be repeated several times until convergence". Could anyone please provide some further information about this?

Thanks,

Xin Fan · 05-24-2012, 10:40 AM

Any ideas?

aaronh · 05-24-2012, 06:46 PM

Things that people have been doing include:
taking the intersection of multiple snp callers
using the hard filters
using the top x percent

By convergence, they mean use one of the above sets as truth, rerun the calibration on all of them, use one of the above methods agains, rerun the calibration and so on until your set of snps stops changing.

Xin Fan · 05-24-2012, 07:01 PM

Thanks, it seems that I have to run it several times and try different thresholds.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SNP function prediction in non-human species	spark	Bioinformatics	7	01-22-2013 02:45 AM
Anyone knows know to access the Human Epigenome Project database..	mandova	General	0	12-20-2011 06:49 AM
ChIP-Seq: hmChIP: a database and web server for exploring publicly available human an	Newsbot!	Literature Watch	1	12-04-2011 03:33 AM
what next after recalibration?	adrian	Bioinformatics	0	12-14-2010 06:08 PM
SNP Database with Solexa/Illumina Reads	foolishbrat	Illumina/Solexa	0	10-05-2009 02:54 AM

05-23-2012, 01:05 PM	#1
Xin Fan Junior Member Location: US Join Date: May 2012 Posts: 3	How to do the recalibration of non-human species without good SNP database Hi all, I am dealing with the genome data of one bird, there are only a few studies about this bird (thus I do not have good SNP data to do the recalibration). I wonder how can I do the recalibration. I found a suggestion from GATK website: http://www.broadinstitute.org/gsa/wik... They said we can bootstrap a database of known SNPs. However, I still have some questions. 1. When I do the initial round of SNP calling on my unrecalibrated data, Should I use any conditions or not? My colleagues are genotyping other genomes. When they do the recalibration (also without good SNP database), they are only requiring that genotypes containing an SNV had a genotype quality score ≥10. If I need to do the same thing as they did, what threshold should I use? They have high coverage genome (>20X), whereas my bird genome is low coverage (15X, after running DepthOfCoverage). 2. They also said "These steps could be repeated several times until convergence". Could anyone please provide some further information about this? Thanks,

05-24-2012, 10:40 AM	#2
Xin Fan Junior Member Location: US Join Date: May 2012 Posts: 3	Any ideas?

05-24-2012, 06:46 PM	#3
aaronh Member Location: California Join Date: Sep 2008 Posts: 45	Things that people have been doing include: taking the intersection of multiple snp callers using the hard filters using the top x percent By convergence, they mean use one of the above sets as truth, rerun the calibration on all of them, use one of the above methods agains, rerun the calibration and so on until your set of snps stops changing.

05-24-2012, 07:01 PM	#4
Xin Fan Junior Member Location: US Join Date: May 2012 Posts: 3	Thanks, it seems that I have to run it several times and try different thresholds.