Quality values for aligned reads in .sam file

[arnkas]

Hi,
I am mapping ABI solid reads using bwa, and then converting to a sam file and ultimately to a .pilup file using samtools. I am trying to understand the quality values in the sam file. For the unmapped reads the quality values agree with those in the fastq file, but for the mapped reads, I can see no relationship between the quality values in the .sam file and those in the .fastq file, or the mapping quality. I bring this up because I am trying to write my own snp filter, and don't know whether I can trust the reported quality values. I would appreciate any help on this.

Arnold Kas

[nilshomer]

Hi,
I am mapping ABI solid reads using bwa, and then converting to a sam file and ultimately to a .pilup file using samtools. I am trying to understand the quality values in the sam file. For the unmapped reads the quality values agree with those in the fastq file, but for the mapped reads, I can see no relationship between the quality values in the .sam file and those in the .fastq file, or the mapping quality. I bring this up because I am trying to write my own snp filter, and don't know whether I can trust the reported quality values. I would appreciate any help on this.

Arnold Kas

It converts the color qualities to base qualities based on the alignment. For example, if neither color was aligned as an error (i.e. corrected), then it returns the sum of the two color qualities plus 10. MAQ and BWA were written by the same author and follow the same color space alignment code to translate the base qualities. See "csmap2ntmap.cc" in maq-0.7.1 lines 102-110 for the exact code.

Briefly, let X and Y be indicators whether or not the first and second colors encoding a base are aligned as errors. Let A and B the color qualities (log probabilities) of the two colors respectively. Then the returned base quality is:

if(1==X && 1==Y) {
  return (A + B + 10);
}
else if(1==X && 0==Y) {
  return (A - B);
}
else if(0==X && 1==Y) {
  return (B - A);
}
else {
  return 0;
}

Since A and B are log probabilities, I would interpret this as a log odds ratio.

[arnkas]

Perfect. Thanks.