Cuffdiff output seems inconsistant

[mattanswers]

(I am transferring this question from the RNA-Seq Forum (where I have deleted it) to this Forum)

I have two sets of time series data. Same conditions, just done on different days. With each set, I have separately gone through TopHat, Cufflinks, Cuffmerge, and Cuffdiff.

Then I hierarchically clustered the 'significant' ('yes' in the significant column) results and found a small but appreciable subset of genes that are present at one time point (say 30min) in one set, but absent in the other set at the same time point.

At a closer look, when the gene is significant it is found only once in the gene.exps table output of Cuffdiff, but in the same time point of the other set where it is not significant the gene appears multiple times, each time with an equivalent locus but different log2-fold change values. Also, at least one of the log2-fold change values is very close to the log2-fold change value found for the significant gene in the other set.

My feeling is that if Cuffdiff did not divide up the gene (multiple entries) it would be significant in that set as well.

Does anyone have an explanation for why the gene would have multiple entries in one set, but only one entry in other another otherwise equivalent experiment ?

[mattanswers]

Some info to clarify the above question

Let me add, that I realize the difficulties of 'significant' genes when you do not have duplicates.

The purpose of my analysis is to get an idea of the consistency of results between the two otherwise identical experiments. In other words to check the consistency of the 'bench' experiments.

The question addressing genes being listed multiple times in the output in one experiment, and only once in the other experiment remains.