HTseq: Very few counts recognised

[sindrle]

Hi!
Ive seen a lot of threads on this, but I can't figure it out. I got 16-60 millions single end reads in each library. Ive used Tophat 2 with UCSC GTF file for hg19.

This is my code:

samtools view accepted_hits.bam | \
htseq-count -m intersection-nonempty -s no -a 10 \
- UCSC/hg19/genes.gtf \
> Out.txt

Here is a typical result, its propotional to the library size:

no_feature 7013689
ambiguous 269370
too_low_aQual 0
not_aligned 0
alignment_not_unique 6645341

How come i get on average 25 - 50% reads that is "no_feature",
"ambiguous" or "alignment_not_unique".

This is RNAseq, and if I must visually inspect, how to precede?

[dpryan]

Perhaps you have a lot of immature mRNAs or a lot of expressed repeat regions. The general idea is to look at some of the alignments in IGV and see if they really don't match anything. Also ensure that the chromosome names in the BAM file and GTF file match (that probably causes this sort of thing half the time).

[sindrle]

Hi again!

I have now tested HTSeq with all modes, also upgraded to Python 2.7.6 and inspected using IGV.



Here is in total 4 reads, one with mapping quality 50 and three with 3.
I used HTSeq option -a 0, so they should been picked up..

All three modes only counts 1 read.. How can this be?

[dpryan]

HTSeq-count also looks at the NH auxiliary tag. With a MAPQ of 3, it's likely that three of those are multimappers (this will be the case if you used tophat2) and would be (properly) ignored.

[sindrle]

Oh yeah, that make sense.

How come you know so much about everything? Where have you learned?

But, its pretty sure something is wrong here right, so I should keep looking? I have checked my GTF, the chromosome names are the same.

[sindrle]

I have made a SAM file with -samout option and checked around a bit..

From Tophat.log I get 21.8 mill. total kept reads
In my SAM I get a total of: 20.90 mill., wonder where the rest are?

Of the 20.9 mill. I have:

17.7 mill. NH:i:1 of which 158.000 ambiguous
I also get 3.4 mill. alignment_not_unique &
1.6 mill. no_feature

Looking at the HTSeq output file I get:

no_feature: 3.8 mill.
ambiguous: 158.000
alignment_not_unique: 3.4 mill.

So the SAM has 1 million reads less than the BAM. Also "no_feature" is different in the SAM and HTSeq output..

I tried to watch specific reads in IGV, but selecting reads by read name (right click the BAM track and choose "select by name", does not change the view....Annoying).

But anyone have something to add on this?