SNP calling on 454 data

bioinfosm · 10-15-2008, 10:55 AM

Anyone has ideas on how to make variation calls on 454 re-sequencing data?

perhaps using the Alldiffs or HCDiffs files from gsmapper software? or some other tools. I believe there needs to be some downstream analysis after Marth lab's mosaik tool, in order to get variation positions and % calls for A C G Ts

cariaso · 10-15-2008, 02:51 PM

I'd hoped we were working on similar things, but it seems not. Your problem seem to be more about recognizing novel snps, which is substantially different from my need to recognized named snps.

Specifically I need to turn the PGP10 exome fasta into a series of dbSNP rs#s and report observed genotypes. Results will be tab delimited and look something like
http://www.snpedia.com/files/prometh...-23andMe-1.csv
Since this is about recognizing named entities, I'd like to extend it to also recognize non-SNP features such as Huntington's, and possibly CNVs.

Sorry I can't be more helpful, but if anyone has code or advice on either topic I'm interested in both.

Tom Bair · 10-16-2008, 08:36 AM

We are working with this, we use mostly the HCDiffs file with alot of post processing. Key things we look at are read depth (hcdiffs is a depth of 3, 2 one way 1 the other)I would say 5 is a better minimum, 15 if you are looking for hets. We also filter for known snps using the dbsnp track from ucsc database and if it is in an exon (also from ucsc) since most people I am working with are looking at nimblegen capture experiments, primarily focused on exons. If you are looking outside exons conservation score appears somewhat useful.

don't know if that helps at all

bioinfosm · 10-22-2008, 08:09 AM

Thanks Tom, that was helpful.

Any others looking for SNPs from 454 data? I heard brute blast approach with no gaps also works! lots of try-it-out-yourself

timread · 10-29-2008, 09:42 AM

We are primarily looking for SNPs in bacterial genomes (ie no heterozygotes). For a first look we parse the HCDifs file for differences with >85% agreement. We then proceed to validation. Most of the single base insertions and deletions turn out to be false positives.

Tom Bair · 10-31-2008, 10:05 AM

timread,

Could you give some parameters on read depth for the false vs true positives? Or do you find no correlation.

Thanks

Tom

timread · 11-02-2008, 11:08 AM

No correlation I can see in the differences called by newbler runmapper that we validated (which are generally high quality calls). I dont think we have a large enough sample size though. We have noted trends in the raw output from runmapper for calls that fall underneath our cutoof filter. Like a large number of 1 bp insertions and deletions are <25-fold read coverage and <50% concordance.

tim

Josliu · 11-21-2008, 01:45 PM

You may use NextGENe software to call SNPs using 454 data. The software links the calling to dbSNP database if GenBank format is provided. SoftGenetics may provide a demo to use NextGENe to your own data.

josliu

Layla · 11-25-2008, 04:28 AM

This is quite a tricky process, especially without the support of bioinformaticians. The downstream analysis is much more complex than carrying out the capture array itself. The HCDiffs file does seem very promising for extracting useful information for SNPs.

Tim, could please say what you mean when you say that you parse the HCDifs file for ""differences with >85% agreement"". and also the kind of validation you do? As I am also certain that alot of our indels will be false positives.
(Thankyou)
Has anybody attempted denovo contig assembly from their capture array data?

Layla

timread · 11-25-2008, 10:04 AM

Quote:

Originally Posted by Layla

Tim, could please say what you mean when you say that you parse the HCDifs file for ""differences with >85% agreement"". and also the kind of validation you do? As I am also certain that alot of our indels will be false positives.

Layla

Layla - by '85% agreement', I mean 85% of the 454 reads agree with the variant call. This is the final column on the header line of the HCDifs file. Verification is by Sanger sequencing.

Layla · 11-27-2008, 02:12 AM

Thank you Tim, I realized what you meant 2 seconds after I had posted the question! Yes, I have been focusing on that file and using diffs > 75% agreement. Cheers, Layla

RockChalkJayhawk · 12-14-2009, 01:32 PM

Quote:

Originally Posted by Layla

This is quite a tricky process, especially without the support of bioinformaticians. The downstream analysis is much more complex than carrying out the capture array itself. The HCDiffs file does seem very promising for extracting useful information for SNPs.

Tim, could please say what you mean when you say that you parse the HCDifs file for ""differences with >85% agreement"". and also the kind of validation you do? As I am also certain that alot of our indels will be false positives.
(Thankyou)
Has anybody attempted denovo contig assembly from their capture array data?

Layla

Layla,

Have you found anyone that has done the contig assembly? I'm curious...

Layla · 12-20-2009, 04:30 AM

Nope, Sorry! I am no longer working on this project

Tuxido · 12-23-2009, 03:35 AM

We also use the hcdiffs in combination with our own downstream analysis where we annotate the data with known SNPs and other useful info. Seems to work fine as long as you have sufficient coverage and there's not too many variants close to each other. With lower coverage you start getting more false positives but you also start missing variants. Actually we once did a comparison with a SNP array and the HCDiffs of version 1.0 of the mapper software and that didn't look that good, as we were missing quite a few variants.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SNP Calling for 454 reads	empyrean	Bioinformatics	1	11-17-2011 07:23 AM
SNP calling from transcriptome data	lre1234	Bioinformatics	2	08-26-2011 08:16 AM
SNP calling for captured sequence data	Mali Salmon	Genomic Resequencing	14	05-05-2011 02:21 AM
New Paper: High Quality SNP Calling Using Illumina Data at Shallow Coverage	nmalhis	Bioinformatics	0	03-01-2010 02:40 PM
SNP calling on 454 data	bioinfosm	Bioinformatics	0	10-15-2008 10:53 AM

10-15-2008, 10:55 AM	#1
bioinfosm Senior Member Location: USA Join Date: Jan 2008 Posts: 480	SNP calling on 454 data Anyone has ideas on how to make variation calls on 454 re-sequencing data? perhaps using the Alldiffs or HCDiffs files from gsmapper software? or some other tools. I believe there needs to be some downstream analysis after Marth lab's mosaik tool, in order to get variation positions and % calls for A C G Ts

11-21-2008, 01:45 PM	#8
Josliu Junior Member Location: State College, PA Join Date: Nov 2008 Posts: 4	SNP calling for 454 data You may use NextGENe software to call SNPs using 454 data. The software links the calling to dbSNP database if GenBank format is provided. SoftGenetics may provide a demo to use NextGENe to your own data. josliu

11-25-2008, 04:28 AM	#9
Layla Member Location: London Join Date: Sep 2008 Posts: 58	Capture and beyond This is quite a tricky process, especially without the support of bioinformaticians. The downstream analysis is much more complex than carrying out the capture array itself. The HCDiffs file does seem very promising for extracting useful information for SNPs. Tim, could please say what you mean when you say that you parse the HCDifs file for ""differences with >85% agreement"". and also the kind of validation you do? As I am also certain that alot of our indels will be false positives. (Thankyou) Has anybody attempted denovo contig assembly from their capture array data? Layla

10-15-2008, 02:51 PM	#2
cariaso Member Location: Wageningen, the Netherlands Join Date: Jan 2008 Posts: 31	I'd hoped we were working on similar things, but it seems not. Your problem seem to be more about recognizing novel snps, which is substantially different from my need to recognized named snps. Specifically I need to turn the PGP10 exome fasta into a series of dbSNP rs#s and report observed genotypes. Results will be tab delimited and look something like http://www.snpedia.com/files/prometh...-23andMe-1.csv Since this is about recognizing named entities, I'd like to extend it to also recognize non-SNP features such as Huntington's, and possibly CNVs. Sorry I can't be more helpful, but if anyone has code or advice on either topic I'm interested in both.

10-16-2008, 08:36 AM	#3
Tom Bair Member Location: Iowa Join Date: Oct 2008 Posts: 26	We are working with this, we use mostly the HCDiffs file with alot of post processing. Key things we look at are read depth (hcdiffs is a depth of 3, 2 one way 1 the other)I would say 5 is a better minimum, 15 if you are looking for hets. We also filter for known snps using the dbsnp track from ucsc database and if it is in an exon (also from ucsc) since most people I am working with are looking at nimblegen capture experiments, primarily focused on exons. If you are looking outside exons conservation score appears somewhat useful. don't know if that helps at all

10-22-2008, 08:09 AM	#4
bioinfosm Senior Member Location: USA Join Date: Jan 2008 Posts: 480	Thanks Tom, that was helpful. Any others looking for SNPs from 454 data? I heard brute blast approach with no gaps also works! lots of try-it-out-yourself

10-29-2008, 09:42 AM	#5
timread Member Location: Atlanta, Georgia Join Date: Oct 2008 Posts: 14	We are primarily looking for SNPs in bacterial genomes (ie no heterozygotes). For a first look we parse the HCDifs file for differences with >85% agreement. We then proceed to validation. Most of the single base insertions and deletions turn out to be false positives.

10-31-2008, 10:05 AM	#6
Tom Bair Member Location: Iowa Join Date: Oct 2008 Posts: 26	timread, Could you give some parameters on read depth for the false vs true positives? Or do you find no correlation. Thanks Tom

11-02-2008, 11:08 AM	#7
timread Member Location: Atlanta, Georgia Join Date: Oct 2008 Posts: 14	No correlation I can see in the differences called by newbler runmapper that we validated (which are generally high quality calls). I dont think we have a large enough sample size though. We have noted trends in the raw output from runmapper for calls that fall underneath our cutoof filter. Like a large number of 1 bp insertions and deletions are <25-fold read coverage and <50% concordance. tim

11-27-2008, 02:12 AM	#11
Layla Member Location: London Join Date: Sep 2008 Posts: 58	Thank you Tim, I realized what you meant 2 seconds after I had posted the question! Yes, I have been focusing on that file and using diffs > 75% agreement. Cheers, Layla

12-20-2009, 04:30 AM	#13
Layla Member Location: London Join Date: Sep 2008 Posts: 58	Nope, Sorry! I am no longer working on this project

12-23-2009, 03:35 AM	#14
Tuxido Member Location: Nijmegen, Netherlands Join Date: Jun 2009 Posts: 22	We also use the hcdiffs in combination with our own downstream analysis where we annotate the data with known SNPs and other useful info. Seems to work fine as long as you have sufficient coverage and there's not too many variants close to each other. With lower coverage you start getting more false positives but you also start missing variants. Actually we once did a comparison with a SNP array and the HCDiffs of version 1.0 of the mapper software and that didn't look that good, as we were missing quite a few variants.