Is VCF output mixing the MQ and Phred scores?

[AmitL]

Hi everybody,

I've noticed something peculiar about the MQ and Phred scores noted in the VCF format.

If you check out these formulas of how these scores are calculated, you can see that the so-called MQ score fits better with the Phred formula, and vice versa.

http://genome.sph.umich.edu/wiki/Mapping_Quality_Scores

http://genome.sph.umich.edu/wiki/Phred_scale

What do you think? Could it be that they mixed them up?

cheers,
Amit

[Heisman]

Looking through those they appear correct to me, although I would have worded things differently. The Phred score is essentially telling you the probability that a specific base call is incorrect. The mapping quality score is giving you a metric to describe the probability that the read in question is mapped to the correct place.

[swbarnes2]

Yeah, those are supposed to be measuring two different things.

MQ is a measure of how likely it is that your read has been mapped to the right place. This is in part due to the quality of the read, which depends on the instrument, partially how well your read matches to the best place in the reference, which depends on how close your reference is to the thing you actually sequenced, and on how repetative the genome itself is. If your genome has perfectly repetative stretches that are long enough, you will never know which one actually made the read you are looking at, no matter how brilliantly your instrument worked. If your read has poor quality, and doesn't match the reference well anywhere, your software will do its best, but it will give it a low mapping quality, to indicate that its not totally confident that this really is the right place.

The phred score is independant of the reference. It's all about how clean the raw data off the instrument was, how confident it is that each letter it's telling you is in the read is an accurate reproduction of the DNA molecule it was given.

[AmitL]

Thanks for your replies!

I understand what these scores mean. What I'm trying to say is that it seems like the numbers in the VCF are misplaced. The MQ value is where the PHRED should be and vise versa.

I will explain myself:
Look at the MQ values in the VCF. Positive integers up to 60. This fits perfectly in the PHRED formula, assuming you can evaluate error rate down to 0.0001%. Then you get a 60 score.

The MQ formula doesn't show a very clear potential for an upper bound like this, and so do the alleged PHRED values.

Moreover, I looked into the paper that explains the MQ score and found that MQ values CAN exceed 60.
Check this graph: http://genome.cshlp.org/content/18/1...expansion.html
Full article: http://genome.cshlp.org/content/18/11/1851.long

What do you say?

[AmitL]

I'm also looking at explanations about the PHRED score, and nowhere have I seen a score higher than 60 being mentioned.

[Heisman]

You make interesting points. I'm not sure if the standard PHRED scale quality score applies since for any one variant it will give a consensus quality score and not individual quality scores for each read. I'm very new to using VCF but this might be worth looking into more.

[Heisman]

I just noticed that by default SAMtools mpileup caps the mapping quality score at 60. This can be changed with the -M option. So perhaps that's why it seems weird.