I am new to RNA-seq analysis and wanted to know what is the best approach to filter out significant genes from the cuffdiff output. The issue I have is I don't have replicates so my understanding I can't use p-values, would it be enough to just rely on the log2 fold change? I appreciate your help.
Fold changes without replicates can support a descriptive analysis but inference using the conventional frequentist concept of statistical 'significance' is not possible without a reliable estimate of the variance for each individual transcript in each experimental group. This is true of any differential expression analysis method (eg see https://support.bioconductor.org/p/42706 ), and has been widely discussed in more general forums - eg see http://seqanswers.com/forums/showthread.php?t=20406 and http://seqanswers.com/forums/showthread.php?t=23586
OTOH if you are a Bayesean, you could try http://www.biomedcentral.com/1471-2105/11/564 although zero citations in 4+ years suggests that the author's prior beliefs have proven to be relatively uninformative....
