To account for over-dispersion of count data, we used a negative binomial regression model with robust standard errors.
For the seed transcriptome data, we used a negative binomial generalized linear model to analyze the contributions of G, PE, and their interaction on the expression level of 4,358 expressed genes.
An alternative linear regression on non-transformed data, using a negative binomial error structure, did not lead to better results.
The DESeq package models count data using a negative binomial distribution [ 51].
The DESeq models count data using a negative binomial distribution, with variance and mean linked by local regression [ 61].
Predictors of number of pupae per container were identified using a negative binomial regression as count data were being analyzed.
For differential gene expression analysis with count data using a negative binomial distribution without replication, the DESeq package in R was used [ 41].
Differentially expressed genes between males and females were determined with the DEseq package [ 42], by modeling count data using a negative binomial distribution.
Recently, a new differential expression calling algorithm, edgeR [ 27], has been proposed, which models count data using a negative binomial model that can be regarded as an over-dispersed Poisson model.
Methods have been developed for the calculation of statistical power for RNA-seq data using a negative binomial distribution ([ 27, 28]; however, these calculations address comparisons made within a single factor.
