Sensitivity Analysis
The proposed iDEG procedure makes two assumptions about the data:
1) a functional mean-variance relationship in
RNA-Seq data exists, and 2) the majority of the genes are null genes. Both
assumptions are commonly accepted and used in the literature;
however, the performance of iDEG is unknown when these assumptions are
violated. Therefore, this section examines the sensitivity of iDEG
to these two assumptions.
Robustness of iDEG to Random Dispersions
We simulate RNA-Seq data from the NB distribution with \(\delta_{g}\) drawn
from a uniform distribution \(\text{Uniform}(0.001,0.1)\). In this setup, \(\delta_{g}\) is no longer a function of \(\mu_{g}\). As shown in Panel A of Figure \ref{fig:sensitivity}, all methods perform worse than in previous studies, but iDEG is still best among the four in terms of the highest \(F_{1}\) scores.