Tophat + htseq_count

[EGrassi]

Hello, I'm performing RNAseq analyses and I've stumbled upon some puzzling results.
I aligned some data with tophat2 (default settings) and as long as the results were disappointing (only about 5% of properly paired reads) I changed the -r and --mate-std-dev parameters and gotten to 60% (I know, still not very high). I ran htseq_count on the resulting bam alignments and comparing the two results I see no differences.
Am I missing something? Does htseq_count use the information about properly paired reads or not? By these results I am prone to say no, I will check the code...

[EGrassi]

A quick check on the htseq_count code tells me that it never uses the reads "mate_aligned" attribute and just considers all of the paired reads. Does this seem a strange behaviour only to me? I don't see in any place a check on wheter the two reads fall at a sensible distance to be reliably considered in the counts.

[Simon Anders]

The "mate_aligned" bit in the FLAG field indicates, in my reading of the SAM spec, that an alignment for the mate is given in the SAM file, not that this alignment is considered plausible. If TopHat really changes the mate_aligned field according to the distance, I'd consider this a very odd behaviour. In my opinion, it should set the alignment quality (5th field in the SAM file) to a low value to indicate that an alignment is reported but should not be trusted.

htseq-count, by the way, filters by the alignment quality only if you use the -a option. I guess I should change this to be the default.

[EGrassi]

The "mate_aligned" bit in the FLAG field indicates, in my reading of the SAM spec, that an alignment for the mate is given in the SAM file, not that this alignment is considered plausible. If TopHat really changes the mate_aligned field according to the distance, I'd consider this a very odd behaviour. In my opinion, it should set the alignment quality (5th field in the SAM file) to a low value to indicate that an alignment is reported but should not be trusted.

As long as the samtools flagstat percentage of properly paired reads gotten on the accepted_hits changed setting the -r tophat parameter I believed that the ones reported as not properly aligned were in the sam file but should not be considered as aligned in the analyses.

(filtering on quality only with an option is fine in my opinion by the way ).