Mapping quality value

[neha]

Hello everyone,

I just want to know how important is mapping quality value when we align the next generation sequencing data with any of the software like bowtie, bwa or bfast. I mean if reads have mapping quality value of 60 or say 50 should be discard them or take them. How should be decide that up-to what mapping quality we will take the reads.

As far as I know mapping quality indicates the confidence that a read mapped to a particular position in the genome is correct.

Please give your inputs.

Thanks,
Neha

[mbblack]

Mapping qiuality is on a log scale, like Phred base Quality scores. So a mapping quality score of 50 is actually equal to an expected error of 1 in 100000, or a mapping accuracy of 99.999%.

I normally set my mapping quality minimum value cutoff at 10, which equals an expected error of 1 in 10 or a mapping accuracy of 90% (I use LifeScope 2.5.1 for our ABI data, but the priniciple is the same for any of the aligners that report mapping quality as the log of the probability).

For bowtie2, just have a look at the manual for guidance

http://bowtie-bio.sourceforge.net/bowtie2/manual.shtml#mapping-quality-higher-more-unique

[neha]

Thank you so much mbblack for explaining it in such a nice way.

[htetre]

Thank you for your explanation, it is very helpful. I am using BWA for mapping, in testing parameters I tend to get a plateau of reads aligned. I was wondering why this is? Is it due to mapping quality? What parameter can I set to decrease the mapping quality so I can get more reads to align to my reference. Thank you for any help you can provide.

[Brian Bushnell]

First off, please note that the mapping score is defined as "Phred-scaled" by the sam specification, but in practice, that's impossible to calculate and the scores vary highly by aligner, as they use different heuristics to estimate. For example, bowtie1 prints quality 99 for mapped and quality 0 for unmapped. To rigorously determine the best quality cutoff, you should look at ROC curves for that aligner and determine which point has an acceptable true positive to false positive ratio for your application.

BWA increases in time exponentially with higher sensitivity and does not increase the amount mapped much more than the defaults (the plateau, as you noted). If you want something with higher sensitivity, please try BBMap. The sensitivity is configurable with the parameters "maxindel" and "minid", which allow you to set the maximum length of insertion or deletion allowed (default 16000), and the minimum percent identity of aligned reads (default 76, minimum 50). It's very fast. If you want something with higher sensitivity than BBMap you'll have to go with BLAST, which is too slow to use on 3rd-gen libraries.

Please ask if you have any trouble using it. Basically you just unzip it then run:
(to index)
bbmap.sh ref=reference.fasta
(to map)
bbmap.sh in=reads.fq out=mapped.sam

It requires Java.