Roadmap

[nignatiadis]

Here is a quite comprehensive list of features which covers a big chunk of what is out there in regards to multiple testing. I am not sure which of these are within the scope of the package (or if everything is), but maybe it is a starting point. This also brings up the point regarding what and how many dependencies we will eventually bring in (e.g. I think Distributions.jl and GLM.jl or so will be unavoidable eventually). I will add references later.

• More p.adjust procedures (to at least get feature parity with MUTOSS)
• Additional pi0 estimators (Issue pi0 estimators roadmap/ideas #3)
• More general interface for multiple testing procedures which don't have equivalent formulation in terms of adjusted pvalues
• Control methods for other error rates, such as:
  • False Discovery Exceedance (False discovery proportion)
  • k-FWER
  • k-FDR
  • PenalizedFDR
• Standard collection of simulations for benchmarking
  • Worst case situations for different procedures
  • Beta uniform
  • Discrete cases
  • Different correlation/dependence structures
• Tests of global null hypothesis
  • Simes
  • Fisher combination
• Higher criticism
• Weighted hypothesis testing
• Grouped / Stratified hypotheses
  • Group BH
  • Stratified BH
  • p-filter
• local fdr/ tail FDR estimation and interface
• empirical null modelling
• Diagnostic or explanatory plots
• Approaches which start from table like for microarrays
  • Procedures which model and account for correlation
  • permutation/resampling/bootstrap based multiple testing approaches
• Approaches for regression (will need dependencies like Lasso.jl/LARS.jl or GLMNet.jl->
  Probably should be a different package)
  • Knockoff filters
• Old school methods
  • Tukey's honest significant difference