call_cnv output .cnv.bed is empty

[yingchen69]

Hi,

I am trying to use CLAMMS to make cnv calls to my targeted panel NGS data. All steps went well except the last step with call_cnv. My bam files are mapped to ucsc hg19, so all chromosomes are started with "chr". call_cnv just gave a segmentation error message with no detail. So I removed "chr" from all chromosome names in all bed files and re-sorted all bed files. This time the call_cnv did not throw out segmentation error message, but the output cnv.bed file is empty. I attached the models.bed and one sample norm.cov.bed file here.

Thanks a lot for the help!

Ying

clamms_models.bed.txt
WGC085336U.norm.cov.bed.txt

[rgcgithub]

Your files look good to me, and I also get no output from call_cnv for your sample. I noticed that you have only ~5k targets, much fewer than whole exome capture. Before I dig too deep, are you sure that your sample has any CNVs in your targeted regions? It's entirely possible that there are no high-confidnece CNVs, which would produce no output.

If you aren't sure, I would try running more samples or using a sample with a known CNV as a test case. You can try increasing the --cnv_rate parameter (increases sensitivity) and also the --mean_cnv_length parameter (used to define the expected CNV size in genomic distance, which may be different from our default parameters in your target platform) if the default settings do not pick up what you are expecting.

Evan

[yingchen69]

Hi Evan, These are 223 gene targeted panel data from kids born with heart disease, not whole exome. It was reported that there are some CNVs in such kinds patients, and that's why we want to run the CNV analysis on the data set. But we are not sure if there is any CNV in the targeted regions in our patients. Best, Ying

…

________________________________ From: Regeneron Genetics Center <[email protected]> Sent: Tuesday, April 4, 2017 10:08 AM To: rgcgithub/clamms Cc: yingchen69; Author Subject: Re: [rgcgithub/clamms] call_cnv output .cnv.bed is empty (#6) Your files look good to me, and I also get no output from call_cnv for your sample. I noticed that you have only ~5k targets, much fewer than whole exome capture. Before I dig too deep, are you sure that your sample has any CNVs in your targeted regions? It's entirely possible that there are no high-confidnece CNVs, which would produce no output. If you aren't sure, I would try running more samples or using a sample with a known CNV as a test case. You can try increasing the --cnv_rate parameter (increases sensitivity) and also the --mean_cnv_length parameter (used to define the expected CNV size in genomic distance, which may be different from our default parameters in your target platform) if the default settings do not pick up what you are expecting. Evan — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#6 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AGH_foVJuaYktnoNtWVHaSjUnXAGIApCks5rsk7QgaJpZM4MvNMI>.

[rgcgithub]

I'm guessing that CNVs in those 223 genes are rare, so it's entirely possible that your one test sample had no CNVs detected by CLAMMS. Before we assume that's a bug, I would test the other samples in your cohort and see what you pick up. If that yields no results we can look into it further.

Evan

[yingchen69]

Hi Evan, Thanks a lot for the help! Actually I ran all 510 samples and all I got are the same empty output files.The copy numbers of 232 genes in these 510 samples I got by using parts of tools from CLAMMS (for clamms_models), gatk4 (for normalized cov files .tn.tsv), and DNAcopy/CNtools are mostly around 2, which are consistent with our expectation.

summary(as.numeric(as.matrix(data[ , c(5:514)])))

Min. 1st Qu. Median Mean 3rd Qu. Max. 0.7748 1.9800 1.9970 1.9850 2.0090 3.0530

Best, Ying

…

________________________________ From: Regeneron Genetics Center <[email protected]> Sent: Thursday, April 6, 2017 11:07 AM To: rgcgithub/clamms Cc: yingchen69; Author Subject: Re: [rgcgithub/clamms] call_cnv output .cnv.bed is empty (#6) I'm guessing that CNVs in those 223 genes are rare, so it's entirely possible that your one test sample had no CNVs detected by CLAMMS. Before we assume that's a bug, I would test the other samples in your cohort and see what you pick up. If that yields no results we can look into it further. Evan — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<#6 (comment)>, or mute the thread<https://github.com/notifications/unsubscribe-auth/AGH_fvz5mZVWLOuNnB4ZMKcNNMAE-VhGks5rtP-pgaJpZM4MvNMI>.

[rgcgithub]

I'm surprised that you see no CNVs in your entire cohort. In fact, if you look at your models file, cols 11-17 provide a breakdown of the number of samples in the training set (reference panel) estimated at each copy number (0-6). You'll see that at most targets, column 13 (diploid state) has ~500 samples implying the full reference panel appear diploid. But there are regions where you'll see a consistent non-zero count in column 12 (het deletion) or column 14 (het duplication). For example, if you look at the known, recurrent locus 1q21.1 (1:147083326-147381445), you'll see you have a sequence of 2.0's in the het deletion column. So you should have two carriers of that deletion in your cohort. (You also appear to have one sample in copy number 0 over ~ all targets, likely a failed sample).

So while you shouldn't base the calls off the numbers in the model file, it at least gives you an idea that there should be a handful of CNV carriers in your cohort. If you aren't picking them up, I'd first try increasing the --cnv_rate parameter in the call_cnv module if you're using the default value (3.0e-8). Your normalized coverage variance looks pretty clean, so you could likely go several orders of magnitude - maybe 5.0e-6 to start?

Good luck,
Evan

[rgcgithub]

PS - Here's a test you should run to make sure you have the right input parameters set.

Using the data you sent, you can force a synthetic CNV to occur in your chrX regions by trying both "M" and "F" for the --sex parameter. It will give you a duplication of X if it's a female, deletion of X if it's male.

./call_cnv WGC085336U.norm.cov.bed.txt clamms_models.bed.txt --sex M
X 31115743 153650115 X:31115743-153650115 WGC085336U DUP 2 196 999 0.972 NA NA NA NA 91 22104 99 -991
./call_cnv WGC085336U.norm.cov.bed.txt clamms_models.bed.txt --sex F
no output