I have 9 experiments (human RNAseq data (control/treatment)),I did RNAseq analysis by CLC genomics,now I want to do differential expression between 2 condition(control/treatment) by R.I need DE for all treated vs all controls samples (I mean I do not want DE for each experiment separately).
DESeq2 is a right choice for my goal?
actually I saw its tutorial but the explanation is for one experiment how can I use it for several experiment?
Yes, it looks like DESeq2 would be a good choice. If the 'experiments' are all done under the same technology, and have been quantified using the same bioinformatic techniques, then you can simply provide DESeq2 with the 18-columned matrix. You could normalise out any technical differences between the experiments (batches) by including a term in your model ~ experiment + condition - if within an experiment there is extra structure (pairing, extra conditions probed,...), add a comment explaining what it is as it might affect the answer. Also, if the experiments were done (or quantified) under different technologies, it may be necessary to reconcile the measurements prior to DESeq2 analysis - again more details would allow us to advise, if that were the case.
I have 9 experiments that each experiment is effect of one drug on one cancer (drugs and cancers are different in each experiment ) but the platform is similar and all of them are Illumina HiSeq 2000 (Homo sapiens). My data from RNA-seq analysis are not normalize.
OK, this is quite important information. So when you say "all treated vs all controls samples", what are you trying to achieve: to find genes that have common magnitude of effect for every drug treatment? That sounds an unpromising approach, and I'd recommend getting in touch with a local statistician who could advise on the pitfalls of such a design. It might well be that your best approach is to analyse each experiment separately, given the heterogeneous nature of cancer types, and reconcile them at the level of genelists. There are all sorts of issues (equality of variances, whether you can take advantage of DESeq's empirical Bayes methodology in a design such as this, ...) which might be better explored with proper consultation.
thank you for a reply.you mean for importing data I should provide a matrix with all of treatment and control samples of all experiments?
Correct - if you give a bit more detail about how one experiment differs from another, we may be able to offer better advice.
I have 9 experiments that each experiment is effect of one drug on one cancer (drugs and cancers are different in each experiment ) but the platform is similar and all of them are Illumina HiSeq 2000 (Homo sapiens). My data from RNA-seq analysis are not normalize.
OK, this is quite important information. So when you say "all treated vs all controls samples", what are you trying to achieve: to find genes that have common magnitude of effect for every drug treatment? That sounds an unpromising approach, and I'd recommend getting in touch with a local statistician who could advise on the pitfalls of such a design. It might well be that your best approach is to analyse each experiment separately, given the heterogeneous nature of cancer types, and reconcile them at the level of genelists. There are all sorts of issues (equality of variances, whether you can take advantage of DESeq's empirical Bayes methodology in a design such as this, ...) which might be better explored with proper consultation.