This package implements AdaPTGMM, a flexible multiple testing method that uses arbitrary covariates to model the local false discovery rate.
The main method adapt_gmm() allows flexibility in covariates, testing type, input, and classification model. We include default classification model implementations using neural networks, multinomial logistic regression, glmnet, and generalized additive models.
Users may use their own custom beta models, please read the vignette for more details.
# install.packages("devtools")
devtools::install_github("patrickrchao/AdaPTGMM")
Optionally, can also build the vignette.
# install.packages("devtools")
devtools::install_github("patrickrchao/AdaPTGMM",build_vignettes = T)
We create simulations similar to Section 4.2 of the paper.
# Load package
library("AdaPTGMM")
library("splines")
# Generate data
n <- 3000
x <- rnorm(n)
expit <- function(x) 1/(1+exp(-x))
pi1 <- expit(2*(x-1))
H <- 1 * (runif(n) < pi1)
theta <- H * rlogis(n, location=2, scale=1)
z <- rnorm(n, mean=theta)
# Two sided testing
pvals <- 2 * pnorm(abs(z), lower.tail = FALSE)
# Run adapt_gmm
x <- data.frame(x = x)
formulas = paste("splines::ns(x,df=", c(2,3,4), ")")
alphas <- c(0.01,0.05,0.1,0.2)
# Can use p-values
res <- adapt_gmm(x=x,pvals=pvals, alphas=alphas,
beta_formulas = formulas, model_type = "neural", nclasses= c(3,4))
# Can use z-values
zres <- adapt_gmm(x=x,z=z, alphas=alphas,
beta_formulas = formulas, model_type = "neural", nclasses= c(3,4), testing="two_sided")
# Fitted Gaussian means
print(paste(c("Means:", zres$params$mu), collapse=" "))
# Fitted Gaussian variances
print(paste(c("Variances:", zres$params$var), collapse=" "))
# Selected beta formula from model selection
print(paste("Selected formula:", format(zres$args$beta_formula), collapse=" "))
# Selected number of classes from model selection
print(paste("Selected number of classes:", zres$args$nclasses, collapse=" "))