modtest_ic: Moderated Statistical Tests for Influence Functions

In biotmle: Targeted Learning with Moderated Statistics for Biomarker Discovery

Description

Performs variance shrinkage via application of an empirical Bayes procedure (of LIMMA) on the observed data after a transformation moving the data to influence function space, based on the average treatment effect parameter.

Usage

modtest_ic(biotmle, adjust = "BH", pval_type = c("normal", "logistic"), ...)

Arguments

biotmle	biotmle object as generated by biomarkertmle
adjust	the multiple testing correction to be applied to p-values that are generated from the moderated tests. The recommended (default) method is that of Benjamini and Hochberg. See topTable for a list of appropriate methods.
pval_type	The reference distribution to be used for computing the p-value. Those based on the normal approximation tend to provide misleading inference when working with moderately sized (finite) samples. Use of the logistic distribution has been found to empirically improve performance in settings where multiple hypothesis testing is a concern.
...	Other arguments to be passed directly to limma::topTable.

Value

biotmle object containing output from limma::lmFit and limma::topTable

Examples

library(dplyr)
library(biotmleData)
library(SuperLearner)
library(SummarizedExperiment)
data(illuminaData)

colData(illuminaData) <- colData(illuminaData) %>%
  data.frame() %>%
  dplyr::mutate(age = as.numeric(age > median(age))) %>%
  DataFrame()
benz_idx <- which(names(colData(illuminaData)) %in% "benzene")

biomarkerTMLEout <- biomarkertmle(
  se = illuminaData[1:2, ],
  varInt = benz_idx,
  parallel = FALSE,
  g_lib = c("SL.mean", "SL.glm"),
  Q_lib = c("SL.bayesglm", "SL.glm")
)

limmaTMLEout <- modtest_ic(biotmle = biomarkerTMLEout)

biotmle documentation built on Nov. 8, 2020, 5:10 p.m.