Annotating and Cataloging CNVs in Varseq – Webcast Q&A

We love when our viewers send questions in during the webcast but unfortunately we can’t answer all of them during the time allotted!  If you asked a question see below for answers, or if after viewing, you have any questions that weren’t asked, please feel free to send those over to support@goldenhelix.com.

Does this work for FFPE derived DNA or ctDNA?

While we have never tested the algorithm on FFPE data, there is no reason that our method would not be applicable. For future work, we intend to evaluate our method on FFPE data to determine the algorithm’s performance in this context.


For calling CNVs from targeted panels, what depth is required? How many controls are recommended?

We recommend 30 control samples and an average read depth of 100x.


How did you come up with 100x coverage as minimum requirement for CNV analysis?

We have evaluated the algorithm on a number of datasets, however we have not been able to test the method on any exome or gene panel data with fewer than 100x coverage. The algorithm performance may be acceptable at lower levels of coverage, but without empirical verification, we cannot know for sure.


How does your CNV calling algorithm compare to CNVkit and GATK4? TPR/FDR? Any ROC analysis?

We have compared our method to a number of competing algorithms on both exome and gene panel data in terms of sensitivity and precision. That comparison has been published here: Detection of CNVs in NGS Data Using VS-CNV. We have not compared our method against CNVkit or GATK4, but these algorithms may be included in future publications as our research continues.


Can CNV calls from an external tool kit be uploaded and annotated in VarSeq?

Yes. This feature will be added for the next release of VarSeq, which should be available soon.


How do you distinguish a de novo or inherited CNV?

Currently, the best way to do this would be to use GenomeBrowse to plot the CNVs present in the mother and father alongside the CNVs in the proband. Then you could look at each individual CNV and visually determine which events are present in the mother or father.


If I already have SNP arrays that have CNV capability (e.g., the new Affymetrix Precision Medicine array) can we combine the NGS data (which we may have on only a subset of our sample) with the SNP array data to get an even better grasp of CNV?

The best way to do this would be to convert the SNP array calls to an annotation track using the VarSeq convert wizard. Then, you could plot your SNP array calls alongside your NGS calls to compare the two.


What is the performance with 30X Whole Genome coverage?

We have tested our method on shallow whole genome data with coverage of around 0.05x and have been able to reliably call large events spanning multiple bands of the corresponding chromosome. I would suspect that smaller events would be detectable with greater coverage, but without an empirical analysis it is impossible to know for sure.


Importing CNVs from VCFs seems to be available now. When will exporting be implemented?

We don’t have a specific timeline on this feature, but it is something that we are looking into.


What’s the input file? Assume it’s bam files. Also, is this for full gene deletions or just partial gene CNVs. Can it detect homozygous CN=0 in genes? I’m not familiar with VS, so is this a brand-new functionality?

For input we use BAM files for coverage data, and VCF files to extract allele frequency data. Our method can detect both partial and full gene deletions, including both homozygous and heterozygous deletion events. This functionality has been around for around two years, but we have made several improvements since first releasing CNV support.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.