All is a bit arbitrary, of course, so maybe I just mean "all within a certain alignment quality boundary". I've wrestled with this time and time again - it seems that no matter how I tweak an aligner's settings I can always take the unaligned reads and run through another aligner, against the same reference, and find more alignments. Often there are really good alignments in that results so it's not just that one aligner can find alignments with more errors that the first failed to find.
Maybe even more of an issue is with certain multi-mappable reads. I might have one aligner only give me a few alignments while another gives me dozens. I know how to use aligners well enough to understand how to control multi-mapping so I'm sure it's not just because I have them set differently. More often than not I want to know if a read maps to multiple loci or not and if so does it map to A LOT of loci to the point that it's a read that I want to discard.
The best solution I can come up with is to use a consensus type of approach. Run multiple aligners, maybe even with different settings (for instance running BBMap with '-slow' and '-vslow' picks up additional alignments that it misses when using mostly standard settings), pooling said alignments, remove redundant alignments, and move on from there. I really don't want to do that due to the amount of time it will take running all of those alignments but it seems like there's just no single pass solution that works.
Comments?
Maybe even more of an issue is with certain multi-mappable reads. I might have one aligner only give me a few alignments while another gives me dozens. I know how to use aligners well enough to understand how to control multi-mapping so I'm sure it's not just because I have them set differently. More often than not I want to know if a read maps to multiple loci or not and if so does it map to A LOT of loci to the point that it's a read that I want to discard.
The best solution I can come up with is to use a consensus type of approach. Run multiple aligners, maybe even with different settings (for instance running BBMap with '-slow' and '-vslow' picks up additional alignments that it misses when using mostly standard settings), pooling said alignments, remove redundant alignments, and move on from there. I really don't want to do that due to the amount of time it will take running all of those alignments but it seems like there's just no single pass solution that works.
Comments?
