DESeq with time points & different samples

[piemmea]

Hi all,
I'm a PhD student newbie to statistical analysis in RNAseq. I got an experimental design and I'm trying to find a good way to approach the analysis of the results.
Basically, I have 5 different fruit varieties harvested at 4 different time points that have been RNAsequenced (so far with no replicas, don't know if we'll get a second shot). I was asked to find those genes showing a different behavior in the 5 varieties both in terms of differential expression on the respective time points and in terms of time course. I thought to use DESeq with the "blind" method (according to the vignette) by making a pair-wise comparison of one of such varieties (and in one time point) with the others (and the other time points, within the same, repeated for the 5).

The problem is, that I am not sure whether this is the most straight-forward way to approach this problem and, moreover, it is still not so clear to me how to represent the results (if making a mega-merge or so).

Does anyone have/can give me a hint?
many thanks
Marco

[dietmar13]

not so easy...

I don't want to offend you - not al all - but perhaps your supervisors...

why do people plan their experiments not carefully? nobody would make a western blot without knowing how to analyse and interpret the results, but thousonds of bucks get invested in NGS and there is no plan who will and how to analyse.

For you design I see no satisfactory solution. without knowing at least a littlebit about biological variance of each sample and timepoint there is no chance to get robust differentially expressed genes.

If I would have 20 samples free, I would have analysed e.g. either the 5 fruits with 4 biological replicates or 2 fruits at 3 timepoints with three replicates.

One chance will be a gene (set) enrichment analysis (G(S)EA) which works different (it accumulates statistical information over genes, not over samples)) and does not need necessarily biological replicates.

I don't know if a time-series version of G(S)EA is available, at least a version for multigroup analysis is available, implemented in the R-package limma (mroast-function). you can import your count-table (htseq-count) with voom() create a contrast matrix and analyse with mroast.

[piemmea]

Thank you Dietmar, I'll try with that.
Regarding the general thing, we are interested in spotting at differences in ripening, and thus at genes that are differentially expressed with each other during the process.
I thought I could first look at the time development and then, after having identified (grouped) those genes showing the same expression behavior, perform a kind of pair-wise comparison with one variety that I think I can
surely take as reference.
Maybe this is not optimal, but is the best thing I can think about at the moment. Do you think is reasonable?
Thanks