Tweet
I'm working with two replicates each for a two group pairwise comparison of single-end RNA Seq data derived from Illumina TruSeq v2, with fragments of ~50 bp.
I've aligned all samples using STAR and Gencode's mouse M8 primary assembly and annotation data, with an average of 88% reads mapping uniquely across all 4 samples.
I am now considering the different approaches for quantification. In the past, I've used feature counts with the default -t exon and -g gene_id flags.
However, when using -t exon and g-gene_id, genes that contain more than one exon have a count summary like this:
Gene Chr Start End Strand Length Count
Gm1992 chr1;chr1 3466587;3513405 3466687;3513553 +;+ 250 6
One issue I'm wondering about is how RPKM will be accurate considering the exon are of different length (first exon here is 100 bp, and second is 150 bp). At this point, I no longer know how many of the 6 reads mapped to each exon, etc. Doesn't this cause a bias?
I considered using -t gene rather than -t exon, but this will include intron regions and also skew the analysis. I've seen some papers discussing significance for intron mapping in RNA seq and nascent transcription, but I don't think this experiment requires this knowledge.
Using -t transcript would also be something to consider, but the reads are single end and small, and the overlap issue is another thing I need to consider.
Can someone give me an idea of pros and cons to using either, and what priorities to consider?
I plan on doing a separate DE analysis using just exon to examine alternative splicing, but I want to do a standard gene/transcript DE analysis first.
What can I do on my end to prevent any more additional noise and bias in this analysis?
Would you guys recommend htseq over featureCounts, or another tool instead?
I usually use edgeR after generation raw counts, but I'm open to hearing about other options.
Thank you so much!
I've aligned all samples using STAR and Gencode's mouse M8 primary assembly and annotation data, with an average of 88% reads mapping uniquely across all 4 samples.
I am now considering the different approaches for quantification. In the past, I've used feature counts with the default -t exon and -g gene_id flags.
However, when using -t exon and g-gene_id, genes that contain more than one exon have a count summary like this:
Gene Chr Start End Strand Length Count
Gm1992 chr1;chr1 3466587;3513405 3466687;3513553 +;+ 250 6
One issue I'm wondering about is how RPKM will be accurate considering the exon are of different length (first exon here is 100 bp, and second is 150 bp). At this point, I no longer know how many of the 6 reads mapped to each exon, etc. Doesn't this cause a bias?
I considered using -t gene rather than -t exon, but this will include intron regions and also skew the analysis. I've seen some papers discussing significance for intron mapping in RNA seq and nascent transcription, but I don't think this experiment requires this knowledge.
Using -t transcript would also be something to consider, but the reads are single end and small, and the overlap issue is another thing I need to consider.
Can someone give me an idea of pros and cons to using either, and what priorities to consider?
I plan on doing a separate DE analysis using just exon to examine alternative splicing, but I want to do a standard gene/transcript DE analysis first.
What can I do on my end to prevent any more additional noise and bias in this analysis?
Would you guys recommend htseq over featureCounts, or another tool instead?
I usually use edgeR after generation raw counts, but I'm open to hearing about other options.
Thank you so much!