GATK Realignment, comparison of the variants between realigned reads and not realigned reads.

Question

[tonja.r]

Hello,

I want to see an impact of the realignment step (GATK) on the variants in the variant calling. So, I have variants of the realigned reads (A) and variants of the not realigned reads (with out a realignment step) (B). In order to see the difference between variants I build up differences A-B and B-A according to the positions of the variants.

My idea is that B-A difference will return me possible false positive variants. What appears in B and does not appear after a realignment in A should be caused by not detected indels in the reads and lead to the mismatches and after a realignment step it got fixed and the reads are shifted so that there are no mismatches anymore.

I guess I am pretty right about it B-A difference but I have problems with interpreting A-B difference. I thought that it could return me true positives that could be detected only after shifting the reads (after he realignment step). I have looked in IGV at the positions of A-B difference and have found that there were as well false positives, so that the reads got shifted and the were no mismatches in the alignment anymore. I guess my idea about A-B difference and true positives is wrong, isn't it?

Thanks in advance.

Answer 1

[Devon Ryan]

I'm not surprised that the realignment step isn't perfect. You'll find it to be a good practice to filter out SNPs neighboring InDels for this very reason. I'd recommend doing that at least for the first pass of an analysis (i.e., if you don't find anything causative at first, then perhaps the filtering removed the causative variant).

[tonja.r]

Well, it is my assignment for my classes and that is the only thing I have to do and to analyze the results. Evene theoretically. So, I have only those two theories about B-A and B-A difference. Is the A-B theory with true positive correct?

[Devon Ryan]

I would say "more likely to be true". I'd be very hesitant to assign results to categorical "true positive"/"true negative" and so on groupings.