Tophat: how to define primary alignment

[qinbird]

From the sam file generate by Tophat, if there are multiple mapped reads, one of the alignments will be labeled as primary alignment.
does anybody know how Tophat define primary alignment?
is there any reference mentioned it?
thanks!

i checked one read, there are four alignments.
all of them have 2 mismatch, but only one are labeled as primary alignment.

[fongchun]

Did you get ever figure this out? I would be interested in knowing this answer.

Thanks,

[dvanic]

I am also quite interested in this! Still...

[dvanic]

Because some of my "non-primary" alignments look quite good in IGV...

[apredeus]

There seems to be confusion in what's meant by "primary" in the field.

E.g. SAM flag interpreter http://picard.sourceforge.net/explain-flags.html clearly uses "non-primary" as an equivalent for spliced

[Brian Bushnell]

There seems to be confusion in what's meant by "primary" in the field.

E.g. SAM flag interpreter http://picard.sourceforge.net/explain-flags.html clearly uses "non-primary" as an equivalent for spliced

I don't see the word 'spliced' anywhere on that page. The difference between primary and non-primary is that only one primary alignment is allowed. It was pretty clear, at least until 'supplemental alignments' and made things murky. The intent of supplemental appears to be for representing chimeric alignments, and thus it could be used for splice events, while secondary shouldn't be used that way.

[apredeus]

I don't see the word 'spliced' anywhere on that page.

Flags of 256/272 in splice aligners' output usually means gapped (spliced) alignment to + and - strand respectively. That's what they call "not primary alignment" on that page.

The difference between primary and non-primary is that only one primary alignment is allowed.

Not sure what that means? I figured that if there's one best alignment score the read would be considered as mapping to one genomic position, while if there are few scores that are equally high would be a multimapper. Now, most RNA-seq aligners do allow multimappers up to a certain number of locations, e.g. Tophat's default is 15.

If you call the read that has the best score primary, it's not clear what's the use of having a term like that since others would be discarded anyway?

[Brian Bushnell]

If you call the read that has the best score primary, it's not clear what's the use of having a term like that since others would be discarded anyway?

Depends on the application; sometimes they are discarded, sometimes not - they can be used fractionally when calculating coverage, for example. The most important case is when there is no single best-scoring location. Then one location can be assigned 'primary' and the rest 'secondary'. This allows a tool (or person) aware of secondary alignments to use that information later, or for simplicity (for example, when converting sam to fastq), all the secondary sites can just be ignored.

[apredeus]

You're right, I was actually totally wrong to assume that flag to mean spliced as opposed to non-spliced alignment. It's "secondary" as defined by some other criteria.

Thank you

[dpryan]

Flags of 256/272 in splice aligners' output usually means gapped (spliced) alignment to + and - strand respectively. That's what they call "not primary alignment" on that page.

That's not splicing, it's a chimeric/fusion/non-linear alignment (there are a bunch of terms for this). These days, those should have the 0x800 bit set in the flag.