Mapping Quality Scale Filtering

Question

[rob234king]

I have a highly repetitive, 3x versions of chromosome containing reference with a great deal of similarity to map to.

I've mapped reads using bowtie2 but want to exclude those of reads that could have mapped to two or more locations so were not identified a unique position. I was thinking best way was to filter SNP calling i.e. samtools mpileup based upon mapping score as the better the mapping the higher the score? But I don't know what value of mapping score (-q) to start filtering from because I don't know the scale or any experience where to start from. Any advise helpful.

Thanks

Answer 1

[Ashutosh Pandey]

When you used bowtie2 you should have gone for the stringent mode. In case of an aligner like BWA, if you have a mpileup then I would suggest to go with a mapping score of 20 that will correspond to 1 in 100 false positives. But Bowtie2 doesnt produces smooth mapping quality scores and the only possible values are

255 = unique mapping

3 = maps to 2 locations in the target

2 = maps to 3 locations

1 = maps to 4-9 locations

0 = maps to 10 or more locations.

I would suggest you to just go with 255 or uniquely mapped reads. I am not sure how much data you would loose like this but you can try. At least it will give you very less false positives.

[Devon Ryan]

You're thinking of tophat2, bowtie2 produces smooth MAPQs.

[Ashutosh Pandey]

Yup you are right. My bad. In that case going with a MAPQ of 20 will be better. Though this is not the best approach but it should be fine.

[rob234king]

So 20 for Bowtie2 and BWA?

[Ashutosh Pandey]

I will go with 20 for both.