Help with HTSeq read-count numbers, a lot of ambiguous or no_feature

[Bernadeta]

Hi, I’m quite a novice to read alignment and mapping and I was hoping someone could advice me on this:

Experiment:
I’ve downloaded sequencing data from a published paper and want to align the reads to the reference human genome. The sequencing data is on the the RNA that was obtained by pulling an RNA-binding protein from an infected cell lysate, treating it with RiboZero and making the library with TruSeq Stranded Total RNA Library Prep. Sequencing was done on MiSeq with 150 single-end reads.

I want align the sequencing reads not to the organism that was used to infect the cells but to the cellular genome itself and to get read counts for the genes the sequences align to.

What I did:

1. Did an alignment with tophat2:

tophat2 -p 16 --min-anchor-length 5 -g 1 --library-type fr-firststrand -o

Align summary
Reads:
Input : 3773481
Mapped : 3624010 (96.0% of input)
96.0% overall read mapping rate.


2. Then I used HTSeq on accepted_hits.bam file to get the read counts for hg38 reference genes:
htseq-count -f bam -r name -s reverse -t exon -i gene_id -m union

I got, however, a lot of reads that fall into no-feture (34%) or ambiguous (37.8%) categories and I’m not sure if this is normal or if I have done something wrong.

Also, the reads with greatest count numbers in HtSeq are still various ribosomal RNAs (top one is 18S with 447001 counts) and I’m not sure if this is normal even after the RiboZero treatment.

Would be immensely grateful for any help.

[fanli]

It is pretty normal to still have some rRNA left over even after a ribodepletion step. Since you are only counting exon-mapping reads, all of the intronic reads (which are not insignificant in a non polyA+ library) fall into 'no-feature'.

'Ambiguous' means a read hits multiple annotated features, but my personal preference is not to ignore these. featureCounts has an option to count these, and is orders of magnitude faster than htseq-count, fyi.

Also, if this was a CLIP/RIP experiment, you may have even less efficiency with the ribodepletion depending on how rough the IP conditions were.