[raonyguimaraes]

You are right, without using this bed file I was getting something like 2 million variants ...

[Heisman]

Yeah, I haven't used GATK before so I can't really say but that seemed like the most logical thing to me (we see hundreds of thousands of mutations prior to filtering just the CCDS regions).

[raonyguimaraes]

On the Unifier Genotyper I'm using the following parameters:

# # #Standard Raw VCF
java -Xmx15g -jar $GATK_DIR/GenomeAnalysisTK.jar -T UnifiedGenotyper \
-l INFO \
-I $OUT_DIR/exome.real.dedup.recal.bam \
-R $REFERENCE \
-B:intervals,BED $EXON_CAPTURE_FILE \
-B:dbsnp,VCF $DBSNP \
-glm BOTH \
-stand_call_conf 50.0 \
-stand_emit_conf 20.0 \
-dcov 300 \
-A AlleleBalance \
-A DepthOfCoverage \
-A FisherStrand \
-o $OUT_DIR/exome.raw.vcf \
-log $LOG_DIR/UnifiedGenotyper.log \
-nt 4

The company where this where done guarantees 30X of coverage ... (http://www.otogenetics.com/human_exome_page.htm)

I know this number should reduce after Variant Recalibrator ... I just want to know how many variants people are getting on this step.

By filtering out mutations you mean using the BED File to call only at the target regions ? If so, yes !

[Heisman]

All right,

Here is my pipeline, some metrics and logs ...

Even after using a bed file from "SeqCap EZ Human Exome Library v2.0" i'm still getting 350191 variants ... Hope someone can help me to get to the promised 20k variants.

I'm planning to switch to GATK 1.2 on the next few weeks ...

Are you sure you are filtering out mutations that are not called in the exome? If you call mutations in the whole genome you will get many more than 20k if you are doing a capture protocol due to non-specific hybridization.

[raonyguimaraes]

All right,

Here is my pipeline, some metrics and logs ...

Even after using a bed file from "SeqCap EZ Human Exome Library v2.0" i'm still getting 350191 variants ... Hope someone can help me to get to the promised 20k variants.

I'm planning to switch to GATK 1.2 on the next few weeks ...

[Heisman]

Papers covering all variations and combinations have been hard to find. I did find one under review (Nature proceedings?) where they claim CASAVA 1.8 comes pretty close to GATK.

I think Novoalign is an excellent aligner, although it requires some tweaking to increase sensitivity on indels that are missed with default settings.

We have done a comparison in our lab with BWA , Stampy, Novoalign, and BFAST. Stampy is the best aligner in our hands (detected more of our SNV and INDEL training set), but Novoalign alignments looked a lot cleaner. I think perhaps with tweaking the gap open penalty for indels, Novoalign might have performed better - just takes some effort to test the parameters more to see if can handle all cases.

GATK is definitely ahead of the game for SNV and indel calling (sensitivity and specificity wise). SAMtools is sufficient - probably you can lean on it if you set the parameters to emphasize specificity instead of sensitivity.

Interesting. Thank you for your post. We do a pretty good job (I think) using the latest SAMtools mpileup command with the -A and -B options and setting a minimum mapping quality per read at 50, but I haven't done anything rigorous to determine what our sensitivity/specificity is. I may go ahead an look at comparing it with GATK.

[raonyguimaraes]

One question. How many raw snps you are getting after running Unifier Genotyper for the first time ?

Here I'm getting about 300 000 snps and I think there is something wrong with this numbers ... Shouldn't it be around 20 000 snps?

I'm running my analysis again using a BED file from SeqCap EZ Human Exome Library v2.0 (http://www.nimblegen.com/products/se...tml#annotation) but still ... 300 thousands snps are a lot ...

[raonyguimaraes]

Thank you so much for posting this pipeline, I've been doing the same for some time. Tomorrow I will post some comments about my results so far.

I think you could sum this pipeline to yours:

https://www.vlsci.org.au/sites/default/files/GatkBestPrac_V3Pipeline_20Sept11.pdf

Let's make from this thread a big reference for who is doing exome sequencing ... Please !!!

[NGSfan]

We use Novoalign + SAMtools... I'm curious if there are any papers out there comparing the methods?

Papers covering all variations and combinations have been hard to find. I did find one under review (Nature proceedings?) where they claim CASAVA 1.8 comes pretty close to GATK.

I think Novoalign is an excellent aligner, although it requires some tweaking to increase sensitivity on indels that are missed with default settings.

We have done a comparison in our lab with BWA , Stampy, Novoalign, and BFAST. Stampy is the best aligner in our hands (detected more of our SNV and INDEL training set), but Novoalign alignments looked a lot cleaner. I think perhaps with tweaking the gap open penalty for indels, Novoalign might have performed better - just takes some effort to test the parameters more to see if can handle all cases.

GATK is definitely ahead of the game for SNV and indel calling (sensitivity and specificity wise). SAMtools is sufficient - probably you can lean on it if you set the parameters to emphasize specificity instead of sensitivity.

[ulz_peter]

Hi guys ,
Thanks for all your responses. I must admit that the GATK parts are a little outdated (already). I'm gonna switch to the new version this week and will update the manual accordingly...

@pc2009open: I can't find any hint for the use of a cluster_file argument in variant quality score recalibration... Anyone else had seen that?

[Heisman]

Great work ulz_peter ! That is exactly what my pipeline is like and I'm glad to see that my choices of tools are also someone else's favorites.

I have seen others use different tools but I think the BWA + GATK + ANNOVAR is the best combination of tools so far...

We use Novoalign + SAMtools... I'm curious if there are any papers out there comparing the methods?

[pc2009open]

Hi ulz_peter,

I have gone through almost the whole process according to your suggestions. However, at the "3.2. Variant quality score recalibration", I encountered some problems. (I used the TATK-1.0.5506 version.)

I got the error message: "Argument with name '--cluster_file' is missing." However, I did not put "--cluster_file" at all.

I looked at some help documents, and found that this kind of "cluster_file" is supposed to be generated by "GenerateVariantClusters". Have you used GenerateVariantClusters before? Is it necessary?

Thanks again for the wonderful manual.

[NGSfan]

Great work ulz_peter ! That is exactly what my pipeline is like and I'm glad to see that my choices of tools are also someone else's favorites.

I have seen others use different tools but I think the BWA + GATK + ANNOVAR is the best combination of tools so far...

[Jon_Keats]

Very nice document, thanks for sharing.

[Heisman]

As the GATK local realignment around indels portion of the website does not explicitly state to "FixMateInformation", I am curious if that will affect downstream analysis in anyway?

Great document, by the way.