bowtie2 parameters for chip-seq

[analyst]

Working with the new bowtie2, anybody here done alignment for chip-seq using it. Please comment of choice of parameters

It is common to choose reads if they match uniquely, for old --best and -m 1 did the trick, not sure about bowtie2, will try -M 1.

[harryzs]

Is there any suggestion about this question?

[harryzs]

From Bowtie2's Manual

it is said:
=======
Mapping quality: higher = more unique

Accurate mapping qualities are useful for downstream tools like variant callers. For instance, a variant caller might choose to ignore evidence from alignments with mapping quality less than, say, 10. A mapping quality of 10 or less indicates that there is at least a 1 in 10 chance that the read truly originated elsewhere.
=======

Could we chose a threshold for Mapping quality for Chip-seq? For example, all reads with

Mapping quality higher than 30 are considered as uniquely mapping reads ?

tks

[harryzs]

I think I find the answer from this paper.

The answer is YES.

"...Reads were filtered by removing those with a BWA alignment quality score less than 15..."

Differential oestrogen receptor binding is associated with clinical outcome in breast cancer
Nature, Vol. advance online publication (4 January 2012) doi:10.1038/nature10730

From Bowtie2's Manual

it is said:
=======
Mapping quality: higher = more unique

Accurate mapping qualities are useful for downstream tools like variant callers. For instance, a variant caller might choose to ignore evidence from alignments with mapping quality less than, say, 10. A mapping quality of 10 or less indicates that there is at least a 1 in 10 chance that the read truly originated elsewhere.
=======

Could we chose a threshold for Mapping quality for Chip-seq? For example, all reads with

Mapping quality higher than 30 are considered as uniquely mapping reads ?

tks

[crepaldi]

What does a mapping quality of 0 mean then? That the read may have originated anywhere in the genome?
And if I understand well, reads with low Mapq should be filtered before calling peaks, right?
Luca

[dpryan]

In bowtie2, a MAPQ of 0 means one of the following:

1. The reported alignment and the next best alignment are both equivalently good, but neither are exact matches (if they're exact matches, the MAPQ is set to 1).
2. The absolute difference in alignment score between the best and second best alignment is >= 10% (and <30%) of the maximum possible difference in alignment scores and the best alignment's score is itself <67% of the maximum difference in alignment scores.

This is for end-to-end alignments. For local alignments, only #1 will produce this (#2 would produce MAPQs of 9, 12, 14, or 17, depending).

Yes, this is highly confusing and no, it's not documented (unless you consider source code to be documentation).

[crepaldi]

Yes, this is highly confusing

Right, then the reasonable way to proceed is to keep the aligned tags with say MAPQ>10 and call the peaks with them. Does it make sense?
Is it possible that the low-scoring tags are still informative, e.g. on the binding to repetitive sequences?

[dpryan]

Right, then the reasonable way to proceed is to keep the aligned tags with say MAPQ>10 and call the peaks with them. Does it make sense?
Is it possible that the low-scoring tags are still informative, e.g. on the binding to repetitive sequences?

That seems like a sensible MAPQ threshold. I agree that the multimappers can still be quite informative. It's likely a good idea to look at them in IGV and bring up a repeatmasker track to see if these might turn out to be interesting or not. The last thing you want to do is throw out multimappers if it turns out that your protein does bind to a repeat region!

[harryzs]

Yes. I agree.
See how people from ENCODE are doing:
samtools view -b -F 1548 -q 30 chipSampleRep1.bam

they(Anshul Kundaje) use -q 30 in their guideline.

https://sites.google.com/site/anshulkundaje/projects/idr