Numerical Studies
Extensive numerical studies were conducted to evaluate the performance of iDEG and to compare it with existing methods, including edgeR \cite{robinson-2007-small-sampl}, DEGSeq \cite{wang-2009-degseq}, and DESeq \cite{anders-2010-differ-expres}, under three experimental settings:
(1) RNA-Seq data follow the Poisson distribution;
(2) RNA-Seq data follow the NB distribution, and the dispersion parameter is a constant; and
(3) RNA-Seq data follow the NB distribution, with a varying dispersion parameter \(\delta_{g}\).
Under each setting, single-subject RNA-Seq datasets are simulated with different
percentages of DEGs, including \(p=5\%,10\%,15\%,20\%\). Each experiment is repeated 1000 times, and for each time, a
baseline transcriptome and a case transcriptome are generated to compose a
RNA-Seq dataset. Performance of the methods are assessed by their
precision, false positive rate (FPR), recall, and \(F_{1}\)
score, which is a harmonic mean of precision and recall. The average
number of identified DEGs are also reported.