What can I do with this data?

[heah]

I have a RNAseq experiment with treatment and control samples of 3 different cell lines (6 samples total) and I want to find out how the treatment changes the gene expressions of the cell lines.

I’ve used DESeq to get a list of differentially expressed genes for each individual cell line, but I’ve learned that without replicates these results are not meaningful. I’ve tried edgeR and CuffDiff as well, but each analysis gives a different list of genes.

Also the cell lines seem too different for use as each other’s replicates as no significant DE genes are found.

I have tried using GSEA and it gives promising results when using the cell lines as 3 replicates, but I’m unsure if it’s correct to use GSEA to analyze RNAseq data.

My questions:
1. Is there something I can do with this data?
2. Are the DE gene lists for individual cell lines worth reporting at all (no biological replicates)?
3. I assume the DE genes can be validated by qPCR. Does each gene have to be validated by qPCR or can I validate x number of genes and say that the whole list is valid?
4. Can GSEA be used with RNAseq data? Is there a guide on how to do this correctly?

[Wallysb01]

I think what you could do is use CuffDiff to call differentially expressed genes from each pair, then look for genes overlapping all cell types. While its not as statistically robust as having replicates, it does give you a kind of validation. Then, you could take those overlapping DE genes and do qPCR as further validation with more replicates.

As far as the number to validate by qPCR, you should do enough to be able to meaningfully say X% validate. Some studies to 100, some 30. Given that you'd need to do this on 3 cell lines, treated and untreated, with at least 3 biological replicates, you're talking 9 samples per gene. So, its probably not realistic to do more than about 30 in your case. That's 270 samples not taking into account technical replicates. If you're going above that, it seems like you might as do more sequencing.

But then you hit a problem with the heavy bias that comes with a different generation of cell, extraction, library prep, and sequencing run. So, if you do expand sequencing in the future, you should be sure to add replicates, incase you find the bias to be so large that you can't combine the two runs.