meassuring interactions between different factor in RNA-seq

[colaneri]

I have and experiment with two genotypes and two treatments, and 4 biological replicates

The most important question to me is whether the genotypes are responding different to the treatments.

I started my analysis using tophat cufflinks cuffdiff but I realized that the tools do not allow me to measure the significance of the interactions. I found that the only way to measure how the two genotypes are performing different in the conditions is using ven diagrams.

X genes from genotype A responded to condition 1. Where X is a list of those genes.

Y genes from genotype B responded to condition 1. Where Y is a list of those genes.

A to measure differential response of genotypes to condition I get those element of X that does not intersect elements of Y.
I don’t like it. If some one know any better way to use the tuxedo tools to perform this kind of analysis PLEASE ILLUMINATE ME.

Now I think that DESeq is a nice tool to perform this kind of analysis.
My question is wheter I can use data already produced with the tuxedo tools as entry table in DESeq.

IF cuffdiff perform differential expression analysis at the level of gene and transcript, then and certain point of the workflow (probably before to compute FPKM) a table with row count per object (transcript gene isoform) must exist. It is possible to get that table. Is that table one of the count tracking files generated by cuffdiff?
It is my proposed approach valid at all?
Althoug similar question have been discussed in the threads I couldn’t deduce an answer to my question from those.

[dpryan]

Yes, DESeq would be a more appropriate tool for you than cufflinks/cuffdiff, since you could then properly model the experimental design. I would encourage you to use counts generated by htseq-count (or similar), since you can then be certain that they were generated correctly. Cufflinks/cuffdiff does produce files with counts, but they're generally either scaled in some way (you don't want that) or they're raw counts that are the sum of transcript counts (it's unclear if that actually corresponds to what you would get with htseq-count or not). So, the conservative route would just be to run htseq-count and then not have to worry about things. With 16 samples, that could be done relatively quickly if you have a computer with a decent number of cores.