effective multimapper filtering

[hans-herma...@mdc-berlin.de]

Hi Alex,

I would like to align short reads to the genome and retrieving only one alignment.

I noticed, if I state i.e. --outFilterMultimapNmax 10, I am retrieving up to 10 alignments per read. Those may be different in their cigar/mapping score, so that STAR either gives me the best alignment , or (if indistinguishable) a "randomly" selected alignment.

Ideally, i would like STAR only give ma the "best" alignment, if there is a indistinguishable "best" among the multi mappers. And if there isn't, than it should remove all of those (bowtie1 has the --best parameter for that). 

I noticed, that even if there is one "best" alignment (i.e. 1x w/o mismatch and 9x with mismatch), the alignment score is the same for all 10. 

The question I have is:

In case i state --outFilterMultimapNmax 1  (not more than one alignment), will cases, in which I have a "best" alignment, be in by output file? Or would it be discarded as being a multi mapper (irrespective of its score)?

best

Harm

  

[Alexander Dobin]

Hi Harm,

STAR's logic is as follows. First it defines how many loci a read maps to, which is controlled by --outFilterMultimapScoreRange that defines the range of alignment score (AS) below the best for accepted alignments.

By default it's =1, so if your best alignment AS=100, the AS=99 alignments (somewhat worse) will also be among the multimappers.

Next, STAR will discard all alignments that map more than  --outFilterMultimapNmax times.

If you want to keep only the best unique alignments, you would need to use  --outFilterMultimapScoreRange 0 --outFilterMultimapNmax 1

Cheers

Alex

On Tuesday, August 9, 2016 at 9:45:38 AM UTC-4,