question about Cufflinks - statistical significance

[caswater]

Hi, I got a question about using cufflinks to determine the differential expression between two samples:

one output is:

MIR941-1 - chr20:62526454-62567384 case ctl OK 1983.51 0 -1.79769e+308 -1.79769e+308 0.46377 0.999964


it is clear that the case sample has gene expression 1983.51, but in the control sample, the expression is 0. I was just wondering why such a discordant expression is assigned with an insignificant p-value here?

Thank you!

[chadn737]

Are there any replicates?

[caswater]

no, i don't have replicates; but some genes are with significant p-vals even if no replicates..

[chadn737]

Those values don't really mean much without replicates. The problem is described well in the DESeq Vignette: "Proper replicates are essential to interpret a biological experiment. After all, if one compares two conditions and finds a difference, how else can one know that this difference is due to the different conditions and would not have arisen between replicates, as well, just due to experimental or biological noise? Hence, any attempt to work without replicates will lead to conclusions of very limited reliability."

In the absence of replicates, Cuffdiff tries to calculates dispersion by pooling the two conditions together (I think it essentially does the same thing DESeq does as Cuffdiff uses DESeqs methodology). DESeq pulls this off by assuming that most genes are NOT differentially expressed. But if your two conditions do differ significantly, then this assumption breaks down.

Without replicates your statistics are not at all reliable and lacks the power to really detect differential expression. Whether or not this is the reason this particular gene is not differentially expressed is the lesser issue and the lack of replication is the real issue.