cuffdiff p value for 2 conditions without replicates

[hug2001]

Thanks masterpiece. It makes sense.
I also read the FAQ on cufflinks website. It says that Cuffdiff compares the log ratio of FPKM in two conditions (i.e. log(FPKMa/FPKMb)) in the following test statistic, which is approximately normally distributed:

T=E[log(FPKMa/FPKMb)] / Var[log(FPKMa/FPKMb)] ~= log(FPKMa/FPKMb) /sqrt(Var(FPKMa)/(FPKMa)^2 + Var(FPKMb)/(FPKMb)^2)

So once we get the statistic we can get a corresponding p value. To get a statistic T, we need to get the Var values.
If there are replicates in each condition, then the Var can be calculated from the replicates;
If there are no replicates, then the Var is calculated under the assumption that the genes/transcripts are not differentially expressed. I guess here the Var is calculated as the Var between the two samples (one in each condition). But definitely Var would be big if the gene is actually differentially expressed in the two conditions and this will make an underestimation of p value under the assumption of "not differentially expressed". So I think the p values would not be precise for no replicates experiments, as masterpiece and mbblack has commented.

More discussion is appreciated if you had such no replicate RNAseq data analyzed with Cuffdiff. How do you feel about the p values?

[masterpiece]

Hi there,

I'm not a statistician, so I can't answer you how the software count the p-value. But I can suggest you reads this thread http://seqanswers.com/forums/newrepl...wreply&p=83270 ( look for mbblack comment #10). He did mentioned bout doing differential expression analysis without replicate which I agree on his opinion.

If you actually have no replicates, then it really is pointless to even bother computing the statistics. In that worst case scenario, you'd do best by simply ranking genes by normalized expression or raw counts, and pick those with the greatest difference in observed values (and then validate them independently).

So you have to interpret your results in light of your experimental limitations, as well as what your goal from the analysis was, and adjust things as the situation calls for. The stats are just tools to guide you and add some rigor to your analysis.

[hug2001]

Hi guys,
Can anyone help explain 1) how p value is calculated for the differential expression using cuffdiff for two conditions without replicates? 2) is this p value useful or we can just look at the fold change to select DE genes?
It is easy to understand p values with replicates, but doesn't make much sense to me to use p or q values for 2 samples without replicates.

Many thanks!