Hello,
I have RNASeq Data of two tissues with 2 replicates each. I processed the data with sleuth and experience a strong asymmetry in the vulcano plot of the differential gene expression.
To see if this is genuine, I implemented the RUVg method in my analysis, which adds an additional parameter in the sleuth fitting process (see below). The values are determined by assuming a set of genes which are not differential between the samples, so the variance they experience is determined by this parameter alone.
RUVg(countData, setOfNormGenes, k=1)
I pass these parameters to sleuth_fit with:
so <- sleuth_fit(so, ~ test1 + condition, 'test1') #with, *so* being the sleuth object and *test1* being a vector of length 4. The first two values correspond to one tissue and the last two to the other.
However, I experience behavior which I can not explain:
A: The result is very sensitive to the presented normalization values. Resulting in the fact that a small variation in the set of genes creates completely different results. As an example (with toy values)
- so$sample_to_covariates$test1 = c(-0.51,-0.5,0.5,0.51) #Looks reasonable in the vulcano plot and the pValues of the DEG are uniform with a peak at 0 (as described here)
- so$sample_to_covariates$test2 = c(-0.5, -0.5,0.5,0.51) #The vulcano plot looks like in the picture below, and the pValues are peaking at 1.
B: For results with reasonable vulcano plots and pValue distributions, the set of genes I set to be non-differential show more extreme beta-Values (see Fig, with beta-values on the y-Axis. No RUV-Normalization on top and the normalization below).
C: (Less of a problem, but still confusing to me)The results are identical when I swap the signs
- so$sample_to_covariates$test3 = c(-0.51,-0.5,0.5,0.51)*-1 # Results test3 are identical to test1
I am currently out of ideas finding a robust way to implement this method without making things even worse, any idea would be highly appreciated.
