giut: Generalized IUT

In LTLA/giut: Generalized Intersection-Union Tests

Description

Perform a generalized intersection-union test for multiple sets of p-values

Usage

giut(..., threshold=0.5)

Arguments

...	two or more numeric vectors of p-values of the same length
threshold	a numeric scalar indicating the p-value threshold with which to estimate proportions

Details

The generalized IUT is a heuristic method that computes (probably) conservative p-values against the union null hypothesis. Consider each configuration of true/false nulls across comparisons, and the proportion of genes corresponding to each configuration. The generalized approach will search across the proportion space to identify the locally maximum p-value.

The landscape is fairly bumpy so convergence is not guaranteed. Rather, optimization is initiated from the most relevant point, i.e., estimates of the proportions obtained using a multi-comparison extension of Storey's method with lambda set at threshold. Maximization will then, hopefully, give a p-value above the true value.

This method depends on a large number of tests to obtain precise estimates of the starting proportions, as well as the values for m1, m2, and so on for each comparison (as described for diut). It also assumes that the alternative distribution is the same within each comparison.

Author(s)

Aaron Lun

See Also

diut

Examples

p1 <- runif(10000)
p2 <- runif(10000)
p3 <- runif(10000)
head(giut(p1, p2, p3))
     
p1[1:1000] <- 1e-8
head(giut(p1, p2, p3))
   
p2[5001:10000] <- rbeta(5000, 1, 20)
head(giut(p1, p2, p3))

LTLA/giut documentation built on May 8, 2019, 7:59 p.m.