I wanted to compare different transcript of the same gene to get an idea which one is expressed more than other.
For this I have 3 samples. I did the alignment using Salmon and then I tximport to pull out abundances.
Can I take a average of 3 samples and make a bar graph to see the transcript expression of a gene?
Yes. Should you? Probably not. There must be better, statistically more robust ways of quantifying based on biological replicates.
Can you please mention any of it. I could not just find any good example from literature as I am only interested in knowing transcript of one gene in one condition (normal)
Like ntsopoul said, explore DESeq2.
DESeq2 is for Differential gene analysis. I also tried DEXSeq (DEU) and DRIMSeq DTU) but my issue is the following:
There is no "differential" then. RNAseq is a relative measure, so comparing single-condition replicates using trivial mathematical operations will be inherently flawed. Fold change metrics make sense, but any normalization or aggregation you do will only serve one purpose: look good in a bar plot; it won't be robust data you can base an investigation on. I deal with data following your exact premise (RNA-seq on 100+ baseline samples, no other condition) and it's difficult to justify any kind of arithmetic we do to say sample X has n times the expression for gene G1 compared to sample Y. We can only say that sample X seems to express more of gene G1 than sample Y.
Thanks, I am not doing any differential. But, Can we say that 6th transcript is relatively more abundance in our sample than other transcripts of that gene?
Oh, you can just use TPM to do that. Within sample comparison is a completely different (and much easier) game.
So can I average 6 replicates??
No. I think you can compare average TPMs as long as you explicitly mention that anywhere you report metrics but don't average counts.
Rob Please, Can you or people from your group answer . As I am using one of your paper as my reference.