Open adamsardar opened 1 year ago
The qqplot of observed p-values against their expected quantiles is incorrect
If we prepare a simulated dataset with known alpha and lambda:
> devtools:: load_all() > simPvals = rbetauniform(1E4, a = 0.126, lambda = 0.391) > betaUniformModel = fitBetaUniformMixtureDistribution(simPvals) > betaUniformModel Beta-Uniform-Mixture (BUM) model 10000 pvalues fitted Mixture parameter (lambda): 0.398 Beta shape parameter (a): 0.126 Negative log-likelihood: -26465.6 > betaUnifPlots = plot(betaUniformModel) > library(patchwork) > (betaUnifPlots$LLsurface + betaUnifPlots$qqPlot)/betaUnifPlots$pValHist
(See top right)
With BioNet:
BioNet
> BioNet::fitBumModel(simPvals) |> BioNet:::plot.bum()
The Beta-Uniform model fits are correct (inspect the parameters), but the qqplots are wrong
There is clearly something wrong with the plot command. There should really be a test against an exact model fit plot too.
What
The qqplot of observed p-values against their expected quantiles is incorrect
Why
If we prepare a simulated dataset with known alpha and lambda:
(See top right)
With
BioNet
:The Beta-Uniform model fits are correct (inspect the parameters), but the qqplots are wrong
How
There is clearly something wrong with the plot command. There should really be a test against an exact model fit plot too.