I have four RNA-seq data from four different samples. Actually I am handling the plant transcriptome data of one control and one treatment for each 2 differential conditions. They have no replicates. So FPKM was calculated for normalization. I want to do DGE between them. I know from another post Gordon suggested that: If FPKM is really all you have, then convert the values to a log2 scale (y = log2(FPKM+0.1) say) and do an ordinary limma analysis as you would for microarray data, using eBayes() with trend=TRUE. Do not use voom, do not use edgeR, do not use DESeq. (Do not pass go and do not collect $200.) This isn't 100% ideal, but is probably the best analysis available.

But It doesn't work for me without replicates. Could you give me an idea to do DGE for this condition? I know It is not ideal without replicates but there is no way for now!

Thanks

But It doesn't work for me without replicates.

What do you mean? Perhaps there are too high expectations, here, for what is a poor experimental design.

I would take Gordon's advice from the other thread. With this experimental setup, one is extremely limited in what one can uncover / find; moreover, one will really struggle to get this work published. Low sample n / low replicate n equates to a low Power to detect the true set of differentially expressed genes. In the very least, one may be able to identify those genes whose expression profiles greatly change between conditions, but, even in this scenario, I would expect that some of these will be false positive associations, which would otherwise not have reached statistical significance with a greater sample n.

The main points of criticism:

• low sample n / low replicate n
• FPKM expression units are not suitable for differential expression analysis

Kevin