Structural Variations With Gasvpro

Question

[int11ap1]

Good afternoon,

I have a .bam file with paired-ends alignments and, with GASVPro, I have predicted some structural variations. However, there is one thing I don't understand. First of all, this software clusters discordant paired-ends. In this step, GASV detects insertions. However, next step is when GASVPro adds the depth coverage information for an improvement in the reliability of the results. Here is when all my insertions disappears.

Why? My theory (maybe wrong theory) is the following one: for novel insertions, there will not be a difference in the coverage of the region I am evaluating and, because of that, there will not be support from the coverage and the insertion previously predicted will be set aside. Nevertheless, if the insertion is n adjacent repetition, there will be a change in the coverage of that region and, does this insertion will be placed into the final results?

I have read the manual but found no clue about it.

Many thanks.

Answer 1

[Charles Warden]

Is the problem only with the insertions?

Try visualizing your sorted .bam files in IGV. It should then be fairly obvious if the coverage supports a predicted indel: even if the reference didn't include the insertion sequence, IGV will show each read spanning the insertion and include those nucleotides in the coverage plot above the reads.

My personal experience is that I haven't been super impressed with discordant paired-end read algorithms for predicting structural variants. For copy number variants, I think coverage works better, and I would recommend CoNIFER. Unfortunately, I don't really have a great alternative for other types of variants, although BreakDancer is probably the most popular (and I've occasionally got potentially useful results that way).

[int11ap1]

Thanks for answering. I'm gonna visualize my .bam files with IGV then. However, I am still curious. I think GASVPro is not able to predict insertions. Is that right? If so, why??? Anybody knows something about it??

[Charles Warden]

I've never used GASVPro, but I would say calling structural variants is often difficult. If you happen to have a set of true positives, my guess is that you will have a hard time recovering them in a reasonably sized list for any program (at least that has been my experience - I would certainly be glad to hear otherwise).

It sounds to me like the program should theoretically be able to predict insertions (otherwise, it wouldn't have tried to look for them at the first step). I would recommend contacting the developers if you have detailed questions:

https://code.google.com/p/gasv/