pi0 estimators roadmap/ideas

[nignatiadis]

Below is a collection of interesting pi0 estimators, which have not been implemented yet. This could also be a good exercise for anyone that would like to contribute!

• ConvexDecreasing (Langaas, 2005), pi0 estimate based on decreasing convex density fit , available in limma: http://www.math.ntnu.no/~mettela/pi0.imf
• FlatGrenander (Langaas, 2005) pi0 estimate based on longest constant stretch of Grenander estimator (don't know of R implementation): http://www.math.ntnu.no/~mettela/pi0.imf
• Cross-validation (leave p-out) based pi0 estimator (Celisse and Robin, 2010) Matlab implementation on http://math.univ-lille1.fr/~celisse/
• CensoredBetaUniform (Markitsis, 2010) , R implementation function CBUM in package pi0
• Central Matching, Empirical Null modeling ala Efron where pi0 is fit jointly with the null parameters
• Restricted MLE Empirical Null modelling, implementation of this or the above would also allow us to provide empirical null functionality
• A pi0 estimator for discrete p-values, e.g. http://arxiv.org/pdf/1410.4274v1.pdf available in fdrDiscreteNull R package
• Estimators in http://arxiv.org/pdf/1204.5488v5.pdf ( implementation: http://stat.columbia.edu/~rohit/Code/NPMixModelCode.pdf )