Cufflink/Cuffdiff how to get fpkm for individual sample

[miaom]

I have 3 groups of samples (A, B and C), and n=4 for each group (12 in total)

Question 1.
Cuffdiff output only gives group-level averaged fpkm. If i want to run some correlation tests for the gene expressions vs. phenotypic traits, how can i get the individual level of fpkm?

Question 2.
If i run cuffdiff in two ways:
- 3 groups altogether (GrpA,B,C) and let the program deal with the statistics (either ANOVA or T-test)
- 3 pair-wise comparisons (GrpA-B, GrpA-C, GrpB-C) one-by-one
I wonder at the end if i could get the same number of significantly expressed genes for each comparison, for example GrpA-B? Did any one test that before?

Thanks in advance!

[andylemire]

Question 1.
Cuffdiff output only gives group-level averaged fpkm. If i want to run some correlation tests for the gene expressions vs. phenotypic traits, how can i get the individual level of fpkm?

The *.read_group_tracking files in the cuffdiff output directory contain the individual FPKMs for each replicate, but I admit I haven't looked at them myself to see how they compare to the group-level average.

Question 2.
If i run cuffdiff in two ways:
- 3 groups altogether (GrpA,B,C) and let the program deal with the statistics (either ANOVA or T-test)
- 3 pair-wise comparisons (GrpA-B, GrpA-C, GrpB-C) one-by-one
I wonder at the end if i could get the same number of significantly expressed genes for each comparison, for example GrpA-B? Did any one test that before?

That depends on how you run cuffdiff, and the three groups themselves. The cuffdiff default uses a pooled model of dispersion across all replicated conditions, so you'd probably get different results from pair-wise versus all together (3 sets of 2 models averaged in the pair-wise, or 1 set of 3 models averaged in the all together). Using the per-condition model of dispersion might overcome this, but I haven't tried it.

[miaom]

thank
You helped a looot!