Fig 1.
Boxplots of estimated feature-specific dispersion parameters of the negative binomial distribution per dataset.
The red dashed line indicates the threshold above which the lack of fit to the negative binomial distribution could not be assessed reliably. See S4 Appendix for details on the datasets used.
Fig 2.
The uniform QQ plots of 500 p-values based on (a) the asymptotic null distribution and on (b) the bootstrap.
The overdispersion parameter is printed on top of each panel.
Fig 3.
Estimated proportions of features with lack of fit to the negative binomial distribution per dataset.
Error bars represent standard errors.
Fig 4.
Estimated proportion of features with significant zero-inflation with respect to the NB distribution.
Error bars represent standard errors.
Fig 5.
Boxplots of proportions of p-values smaller than 0.1 for features with poor and good fit to the NB distribution, per dataset.
The p-values are computed with the Wald test in a NB model, edgeR, DESeq2 and with the Prentice rank sum test.