I've been lurking for a month learning about RNASeq analysis but now it's time to post my first question!
Very briefly I have the following situation that I would love to have some insight about:
RNAseq (HiSeq, single reads, 50bp) on 4 cell lines, untreated or treated.
High-quality 44bp reads after fastqc filtering and trimming, 20-27 million mapped reads on human, tophat2-mapped using hg38.gtf strict.
We originally assumed that the cell lines would be sufficiently similar to each other to be considered "biological replicates" and we would thus see the effects of the treatment on the transcriptome. However, this turned out to be a wrong assumption and they are quite different globally, so the effects of the treatment are far smaller than being different cell lines (I've seen this by clustering normalized reads (from HTSeq) or FPKM values from Cufflinks, and running a PCA or doing heatmaps).
What's the best way to analyze RNASeq without biological replicates?
So far I've tried
Cuffdiff using "-blind" mode, doesn't seem to yield anything different than default setting.
DESeq2, seems to work a bit better but severely underestimates changes (that we already saw by QPCR on individual genes).
So I was wondering if you think it's possible / "allowed" to do fold-difference values on normalized reads (HTSeq and using sizeFactors from DESeq2) pair-wise of untreated and treated of the 4 samples but what statistics can I use for that to deteremine significant changes? Or is it better to use FPKM values from Cufflinks, since they already come with a built-in confidence interval and simply calculate folds pairwise and cluster the results ?
Grateful for any kind of insight or advice!