RNA-seq without replicates, type of analysis to trust.

[santhilalsubhash]

We have two samples control and treatment without any replicates. I want to find differentially enriched / expressed transcripts in treatment over control.
I know it is not good to proceed without any replicates but still there are methods which supports samples without replicates. I just want to know which one is trustable ?

The problem is since I don't have any replicates I can't trust any statistical packages (like edgeR, DESeq or cuffdif) which provides us FDR. I have also used GFOLD which gives logFC and gfoldFC, but I don't know how to fix the cutoff from this fold change values and get the top enriched transcripts. Because if I fix some gfoldFC cutoff and use it for geneontology which does not give any biologically relevant processes for the type of sample I used. For example the sample is pull downed RNA from certain cell cycle phase, where I should atleast expect some cell cycle related terms.
But in case of edgeR or DESeq when I filter with logFC and FDR at-least gives some related terms to the samples.

Or Is it fine if I use GSEA preranked test for all the list of protein coding genes with fold change from edgeR, DESeq or GFOLD?

Is there any analysis where I can trust for samples without replicates. Please let me know how to further proceed with this type of data.

In this paper they have compared the availabe differential expression analysis packages. Where DESeq produces less true positives than edgeR in case of less number of replicates.

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3608160/

[mikesh]

In this case I'd rather use statistical testing for functional enrichment and selected differentially expressed (DE) genes by fold-change. I would select several DE gene lists with e.g. 1.5-, 2- and 4-fold up-regulation and run GO enrichment analysis for them.
Ideally, in your case, GO terms related to cell cycle should be enriched in these sets. Moreover, enrichment odds ratio and -logP-value should be gradually increasing with the increase of DE fold-change cutoff.
IMHO this is one of very few things that could be done in absence of biological replicas.
PS Of course GSEA preranked test should also work fine here

[santhilalsubhash]

GSEA preranked analysis

Thanks mikesh for you suggestion. I am going to use GSEA for preranked list with fold change and lets see whether I can get some meaningful results.

[NicoBxl]

Be very carefull not to make assomption on your results and to force your data to output something. That's not how science works.

[santhilalsubhash]

Yes NicoBxl I aggree with your point. But the problem is I could not get any replicates for the samples. I should go for whatever methods are available and compare the results from all the methods. There is no other choice for me.

Please let me know if you have any suggestion to proceed with the analysis.

[NicoBxl]

you could use DESeq. There is a no-replicate mode.

And of course confirm the expression of several candidates by other methods like qRT-PCR

[pauboher]

I am trying to get some differentially expressed genes from a dataset without replicates...or at least to find some functional enrichment of the transcripts over represented in each dataset...Santhilalsubhash, how are you analysis without replicates are going? Have you find the way to detect functional enrichment of the genes which show higher or lower patterns in comparison to your control?