Bowtie: How to retain only uniquely mapped reads?

[EstherKLather]

Hi!

I have been searching for the answer to this questions on the internet (google, this forum, bowtie manual), but can't find a clear answer.

I am mapping RNASeq reads to the mouse genome . If one read aligns to multiple positions on the genome I want it to not be mapped at all.
I used the --best option for this, but I am not sure if it's right. The bowtie manual only says: "--best: Make Bowtie guarantee that reported singleton alignments are "best" in terms of stratum [...] and in terms of the quality values at the mismatched position(s)." But what happens if two alignments are equally good?

If it is not right to use --best for this purpose, which combination of output options can I use?

Thanks for any help.

Cheers,
Esther

[NicoBxl]

-m 1

exemple on bowtie website :

Example 7: -a -m 3

$ ./bowtie -a -m 3 -v 2 e_coli -c ATGCATCATGCGCCAT
No alignments
Specifying -m 3 instructs bowtie to refrain from reporting any alignments for reads having more than 3 reportable alignments. The -m option is useful when the user would like to guarantee that reported alignments are "unique", for some definition of unique.

[EstherKLather]

Thank you very much, Nico. I'll try with that one.

I am still wondering though, what happens with the output using --best if there are two (or more) best (equally good) alignments.

And am I right if I say that it doesn't make sense to combine -m 1 with -n 2 (allow 2 mismatches), because this would throw out a read entirely if it matches one time perfect and another time with mismatches? I am having a dilemma here, because I want to allow for mismatches.

If --best threw out a read entirely if it has two equally good best alignments and still reported reads that have more than one alignment (including one best) this would solve my problem, wouldn't it? Does --best behave this way?

Cheers,
Esther

[NicoBxl]

yeah indeed allowing two mismatches and only one alignment in the whole genome will not give you good results ( I guess, try it to see the results ).

[fkrueger]

I think the solution to your problem would be a combination of several options, such as -n 2 -l {your_read_length} --best -m 1.

If you do it this way --best is trying to report the best alignment while allowing up to 2 mismatches. If there are more equally good alignments which would be considered 'best', none of them will be reported.

[EstherKLather]

Thanks very much to the both of you!

I now figured out entirely how all the different bowtie output options work together. Sorry for my question, I could have answered it by reading the manual very carefully, but as a starter in this topic I was very confused.

So, thanks again!

[giverny]

Hi there
For human analysis I used -n 2 and -m 1 but I obtained variations within repeat area. What do you think I should do to avoid such results?
Thanks

[giverny]

I forget to say :
My reads are 76 bases long

[giverny]

No idea?
Thanks for your time

[Dario1984]

It is not a good idea to be using Bowtie to map RNA-seq reads. Why aren't you using TopHat ? You are mapping mouse reads and they do have introns, because a mouse is a Eukaryote.

[dhirallin]

I may be mistaken. But I think you also need the --strata option to do what you want. E.g.
-a -n 2 -m 1 --best --strata

You need -a otherwise -m 1 is meaningless (e.g. -k 1 ensures more than 1 alignment will never be reported anyway, so -m 1 will never get evaluated.)
In that case --best by itself just lists all the alignments from best to worst, whereas --strata will cause it to only report alignments with the best strata, then -m 1 can filter out cases where there are more than 1 alignment matching the best strata.