Samtools.1 Parameter Q's

[msleslie]

We are optimizing a custom pipeline that uses some of the functions from Samtools.1 and have a couple of questions about the parameters:

1) Coefficient for downgrading: This makes sense to us as a weighting coefficient that works on the quality scores, but we're a little confused by the formula. Is it correct to multiply the coefficient back to the INT or should it be a multiplier of the quality score (i.e., sqrt((INT-q)/INT)*q)?

From samtools.1 manual (http://samtools.sourceforge.net/samtools.shtml):

-C INT Coefficient for downgrading mapping quality for reads containing excessive mismatches. Given a read with a phred-scaled probability q of being generated from the mapped position, the new mapping quality is about sqrt((INT-q)/INT)*INT. A zero value disables this functionality; if enabled, the recommended value for BWA is 50. [0]

2) Where do the -i and -t parameters get implemented? And we’re not quite sure why setting them is important. Any information would be appreciated.

From samtools.1 manual (http://samtools.sourceforge.net/samtools.shtml):

-i FLOAT Ratio of INDEL-to-SNP mutation rate [0.15]

-t FLOAT Scaled mutation rate for variant calling [0.001]


THANKS!!

[rskr]

We are optimizing a custom pipeline that uses some of the functions from Samtools.1 and have a couple of questions about the parameters:

1) Coefficient for downgrading: This makes sense to us as a weighting coefficient that works on the quality scores, but we're a little confused by the formula. Is it correct to multiply the coefficient back to the INT or should it be a multiplier of the quality score (i.e., sqrt((INT-q)/INT)*q)?

From samtools.1 manual (http://samtools.sourceforge.net/samtools.shtml):

-C INT Coefficient for downgrading mapping quality for reads containing excessive mismatches. Given a read with a phred-scaled probability q of being generated from the mapped position, the new mapping quality is about sqrt((INT-q)/INT)*INT. A zero value disables this functionality; if enabled, the recommended value for BWA is 50. [0]

2) Where do the -i and -t parameters get implemented? And we’re not quite sure why setting them is important. Any information would be appreciated.

From samtools.1 manual (http://samtools.sourceforge.net/samtools.shtml):

-i FLOAT Ratio of INDEL-to-SNP mutation rate [0.15]

-t FLOAT Scaled mutation rate for variant calling [0.001]


THANKS!!

This is interesting. In the benchmarking I have done the mapping qualities haven't been so relevant, because they haven't correlated well with accuracy of the alignments, maybe there is a better way to calculate them.

[clicker]

Regarding "Coefficient for downgrading", I think you may be right. I was perplexed by the following simple examples, where INT=50 (or -C 50) and original mapping quality scores are either (5 or 20). This would result in...

a) mapping quality score of 5 with -C50 ..... sqrt((50 - 5) / 50) * 50 = 47.4
b) mapping quality score of 20 with -C50 ..... sqrt((50 - 20) / 50) * 50 = 38.7

...with new mapping scores appearing inversely correlated.

Also - can anyone define what is meant by "excessive mismatches" from the manual...

-C INT Coefficient for downgrading mapping quality for reads containing excessive mismatches. Given a read with a phred-scaled probability q of being generated from the mapped position, the new mapping quality is about sqrt((INT-q)/INT)*INT. A zero value disables this functionality; if enabled, the recommended value for BWA is 50. [0]


Lastly, I ran mpileup with the same sequences to compare the effect of setting -C50 versus -C 0. The reads had been aligned with Novoalign (not BWA).

When applying -C50, I found that those original reads with low mapping qualities (around 13) stayed as they were (around 13), despite having mismatches. Reads with very high mapping qualities (e.g. 106 or 150), were reduced to about 20 or 25. It should be noted that these reads with high mapping qualities also had paired-ends that either mapped to another chromosome or not at all, which is a bit bizarre given that option -A was "not" set... (where -A = Do not skip anomalous read pairs in variant calling).

When applying -C0, the reads with low mapping qualities again did not change their mapping qualities, whilst those with high mapping qualities (and poorly mapped paired-ends) just disappeared from the mpileup output.