Breakdancer False Positive

Question

[madkitty]

We performed and whole genome sequencing with NGS, and used Breakdancer to detect Copy Number Variation. Breakdancer and config files were made with NO options whatsoever, only with the following lines :

perl bam2cfg.pl input.bam > input.cfg
breakdancer input.cfg > output.txt

The output with Breakdancer is really inconsistent. How can I eliminate false positive ?

[Zev.Kronenberg]

What are you defining as a false positive?

[madkitty]

A false positive is a CNV identified by the software where there is in fact no CNV. It's the same concept as a false positive pregnancy test, it tells you you're pregnant when you're actually not pregnant. Breakdancer has a lot of false positive and aberrant long rearrangement, this is what I'm trying to eliminate. I thought of using other softwares and compare results, but I was wondering if someone has a better method :)

Answer 1

[Chris Whelan]

The output from most SV detection programs can be pretty noisy, but it's usually due to incorrect alignments in your BAM file.

I've found that you can cut down on the noise from Breakdancer a lot by running it with the -g option to write the supporting read mappings for each SV into a file, and then examining the genome around the locations of those mappings. For example, I've found better results if you filter out SV calls where the supporting reads align to annotated segmental duplications (although then you lose sensitivity in those areas).

Another thing to try would be to use the -q option to require higher mapping qualities from the alignments.