minimum MAPQ from bwa mem before peak calling

[ickou]

Hi,

I am working on genomics data (ChIP-Seq, SR100) and for this I used bwa mem (no "XT:A:U" tag) for aligning the fastq files.
After, I am finding peaks with MACS.

It is necessary to use a filter on the minimum mapping quality (MAPQ)...

What is the best way to remove reads that map to multiple locations?
What cutoff value do you use about MAPQ?

There are very different answers depending on the forums...

I think it is only necessary to remove zeros but I don't find a consensus or recommendations for good practice...


thank you !!

Emeric

[Brian Bushnell]

If your reads are shorter than 70bp (which is often the case for chip-seq), bwa's documentation states that it's better to use bwa-aln than bwa-mem.

Ambiguously-mapped reads should have a mapq of 3 or less.

[ickou]

Thank you Brian !

How to justify a threshold of 3 for MAPQ?
Is it an arbitrary threshold or can we find a study on this? I need references (publication, website or other...).

I found only this:
"Non-unique read is placed randomly with a mapping quality 0"

http://www.broadinstitute.org/files/shared/mpg/nextgen2010/nextgen_li.pdf

thank you !!

Emeric

[Brian Bushnell]

Thank you Brian !

How to justify a threshold of 3 for MAPQ?
Is it an arbitrary threshold or can we find a study on this? I need references (publication, website or other...).

That is implicit in the official SAM specification. The MAPQ is the phred-scaled probability that the mapping position is correct. 3 on the phred scale means ~50% chance of being correct. Therefore, a multi-mapping read must have a MAPQ of 3 or lower in a correct implementation.

The presentation you linked is from 2010, and I don't think it reflects bwa's current behavior.