Hi Reena,

One gene ID can be associated with multiple transcript IDs as the gene ID typically denotes genomic region. Given the likely similar functionality of the transcripts and p-values, it is unusual that you would have a transcript that is up-regulated and one that is down-regulated. My suggestion would be to take a geometric mean of the fold changes and store it under one gene ID.

If you are analyzing time-series data, please try out my tool at bit.ly/osudiatool. Any feedback is appreciated.

Thanks,

Austin

Thank You Austin,

Theoretically, I had similar justification for any such discrepancy. Regarding the geometric mean, the variation is 10 folds both in up regulation and down-regulation. Do you still think that using geometric mean would be correct. Mine is just a control and treatment data.

First, I want to understand what steps or tools did you use for DGE analysis.

You must already be aware that one gene can have multiple transcripts (isoforms) which could be expressed differently depending upon the conditions under study.

That being said, say for 2 samples under study (say 1 control and the other treated), it is perfectly acceptable to see 2 different transcripts in 2 different samples. Since, these are just transcripts, they may belong to the same gene which implies that they will have same gene identifier. In your case that will be same RefSeq IDs.

Secondly, do not rely much on the p-value threshold / cutoffs. This is because without having sufficient number of replicates in hand under study, it is meaning less to perform statistical tests. Ideally 3-4 replicates are required.

And, finally coming to FPKM, that is just a normalized representation of the raw read counts. Now, consider 2 scenarios here -

say, you "gene A" under study has 5 exons of varied lengths and is expressed in 2 types to transcripts :

Exon Name Length

E1 5

E2 9

E3 11

E4 3

E5 7

transcript-1 : E1, E2, E4 = length = 5 + 9 + 3 = 17

transcript-1 : E2, E3, E5 = length = 9 + 11 + 7 =27

now, as you can see, because the lengths of the transcripts is not same, the number of reads mapping to the transcripts will differ. FPKM will help normalizing that count based on the varying length. So, if you see different FPKM values for same gene in 2 different samples and if the fold change is large enough then you may think (We are not yet sure, mind it) that gene to be differentially expressed. That's just a clue. Validation is required.

Do share what commands / pipeline or tools you used so that we can discuss this further.

Reagrds

Vijay Lakhujani

I have used TopHat and cufflinks with FPKM normalization. As they have same gene as well as Refseq ID, the validation with qRT will validate only one transcript i guess so.