Alignment QC differences between HISAT2 and Qualimap

Question

I used hisat2 for aligning reads to to the genome. I have an alignment summary for sample1 as follows:

89038751 reads; of these:
  89038751 (100.00%) were paired; of these:
    5641019 (6.34%) aligned concordantly 0 times
    67665552 (76.00%) aligned concordantly exactly 1 time
    15732180 (17.67%) aligned concordantly >1 times
    ----
    5641019 pairs aligned concordantly 0 times; of these:
      651415 (11.55%) aligned discordantly 1 time
    ----
    4989604 pairs aligned 0 times concordantly or discordantly; of these:
      9979208 mates make up the pairs; of these:
        5887822 (59.00%) aligned 0 times
        2656902 (26.62%) aligned exactly 1 time
        1434484 (14.37%) aligned >1 times
96.69% overall alignment rate

To check the alignment quality I used Qualimap. In the report I see the following:

Reads alignment

Number of mapped reads: 172,189,680
Total number of alignments: 244,302,519
Number of secondary alignments: 72,112,839
Number of non-unique alignments:    22,973,098
Aligned to genes:   2,081
Ambiguous alignments:   173
No feature assigned:    178,975
Not aligned:    5,887,822

Reads genomic origin

Exonic: 2,081 / 1.15%
Intronic:   6,866 / 3.79%
Intergenic: 172,109 / 95.06%
Intronic/intergenic overlapping exon:   302 / 0.17%

Transcript coverage profile

5' bias:    0.69
3' bias:    2.85
5'-3' bias: 0.39

Can anyone tell me why the numbers look different? Did I go wrong somewhere? Help me in understanding the results.

Answer 1

89038751 pairs times 2 is 178 million reads. That times the approximately 97% alignment rate is around 172 million reads, which is what qualimap is reporting.

A "non-unique" alignment is one that can map to multiple places (either equally well, or where all of the alignments are of similar enough quality). Such alignments have a single "primary" alignment and one or more "secondary" alignments, which is where the "secondary" metrics come from.

For the qualimap output, there are additional metrics describing how reads overlap genes (or not). Of note are the "ambiguous" alignments, which are those overlapping multiple genes. "No feature" metrics, then, refer to alignments not overlapping any gene (see the qualimap documentation for further details).

My guess is that you have such a high "intergenic" rate because you've either used the wrong GTF file or have told qualimap the wrong strand setting. For unusual organisms one expects many reads to arise from unannotated genes. This is not the case for common model organisms, such as mice and humans, where there are very few remaining commonly expressed unannotated genes.