Discrepancy between HTSeq-count counts and total mapped reads

[M_staats]

Hi All,

There seems to be a rather large discrepancy between the number of mapped Illumina reads and the total number of HTSeq counts (gene counts + counts with no_feature + ambiguous counts + too_low_aQual + not_aligned + alignment_not_unique).

For instance:
I used Tophat2 to map paired-end illumina reads. My two FASTQ files contain a total of 39785174 reads of which 4177881 were unmapped (unmapped.bam). A total of 35607293 reads were mapped (accepted_hits.bam), of which 31390707 were uniquely mapped (grep -w "NH:i:1" accepted_hits.sam | sort | uniq | wc -l).

My input HTSeq-count command was:
htseq-count --mode=union --stranded=no --type=exon --idattr=gene_id accepted_hits.nsorted.sam Solanum_tuberosum.3.0.17.gtf > HTSeq_counts_UNION_ENSEMBL_GTF.out

The HTSeq output is:
Total sum of gene counts 13319749
no_feature 3150201
ambiguous 346347
too_low_aQual 0
not_aligned 0
alignment_not_unique 9024735

The sum of HTSeq-count is 25841032 counts. Shouldn't the sum of counts be equal to the total number of mapped reads (35607293)? Please help me explain this discrepancy.

Thanks in advance for the answer!
Best,

[Simon Anders]

htseq-count counts read pairs, not reads. Maybe this explains it.