I'm analysing some stranded RNASeq data and looking at identifying genuine anti-sense transcripts. What I'm finding with my default STAR / htseq-count pipeline is that htseq-count (quite correctly) reports antisense transcription from those features that overlap, but on opposite strands.
The 3' ends of TLDC2 and SAMHD1 is a good example of such a scenario:
Ensembl region
Sense transcription from SAMHD1 is appearing as antisense transcription from TLDC2 in the htseq-count output. Whilst this is technically correct what I am finding is that if SAMHD1 (sense) is differentially expressed I see TLDC2 antisense as also differentially expressed, but the entire signal is driven by the SAMHD1 overlap, so I do not believe the TLDC2 call is biologically meaningful.
From reading the htseq-count manual again I don't see an option to consider reads mapping to these kinds of overlaps ambiguous. As the union/intersection_strict/intersection_nonempty specification seems to only consider features on the same strand.
So my question is: a) is there some clever way I am missing in htseq-count to consider features on opposite strands to be overlapping (and thereby ignore reads)? and b) if not is there any tool out there that can filter from a GFF file overlapping features and give me a reduced GFF with these regions excluded?
The 3' ends of TLDC2 and SAMHD1 is a good example of such a scenario:
Ensembl region
Sense transcription from SAMHD1 is appearing as antisense transcription from TLDC2 in the htseq-count output. Whilst this is technically correct what I am finding is that if SAMHD1 (sense) is differentially expressed I see TLDC2 antisense as also differentially expressed, but the entire signal is driven by the SAMHD1 overlap, so I do not believe the TLDC2 call is biologically meaningful.
From reading the htseq-count manual again I don't see an option to consider reads mapping to these kinds of overlaps ambiguous. As the union/intersection_strict/intersection_nonempty specification seems to only consider features on the same strand.
So my question is: a) is there some clever way I am missing in htseq-count to consider features on opposite strands to be overlapping (and thereby ignore reads)? and b) if not is there any tool out there that can filter from a GFF file overlapping features and give me a reduced GFF with these regions excluded?
Comment