FDRreg-package: False discovery rate regression

Description Details Author(s) References Examples

Description

Tools for FDR problems, including false discovery rate regression. Fits models whereby the local false discovery rate may depend upon covariates, either via a linear or additive logistic regression model.

Details

Package: FDRreg
Type: Package
Version: 1.0
Date: 2014-02-25
License: GPL (>=3)

The workhouse function is FDRreg(z,X, ...), where z is an observed vector of z statistics, and X is a matrix of covariates. Do not add a column of ones to X to get an intercept term; the function does that for you, just like R's base lm() and glm() functions.

Author(s)

Author: James G. Scott, with contributions from Rob Kass and Jesse Windle.

Maintainer: James G. Scott <james.scott@mccombs.utexas.edu>

References

False discovery rate regression: application to neural synchrony detection in primary visual cortex. James G. Scott, Ryan C. Kelly, Matthew A. Smith, Pengcheng Zhou, and Robert E. Kass. arXiv:1307.3495 [stat.ME].

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
library(FDRreg)

# Simulated data
P = 2
N = 10000
betatrue = c(-3.5,rep(1/sqrt(P), P))
X = matrix(rnorm(N*P), N,P)
psi = crossprod(t(cbind(1,X)), betatrue)
wsuccess = 1/{1+exp(-psi)}

# Some theta's are signals, most are noise
gammatrue = rbinom(N,1,wsuccess)
table(gammatrue)

# Density of signals
thetatrue = rnorm(N,3,0.5)
thetatrue[gammatrue==0] = 0
z = rnorm(N, thetatrue, 1)
hist(z, 100, prob=TRUE, col='lightblue', border=NA)
curve(dnorm(x,0,1), add=TRUE, n=1001)

## Not run: 
# Fit the model
fdr1 <- FDRreg(z, covars=X, nmc=2500, nburn=100, nmids=120, nulltype='theoretical')


# Show the empirical-Bayes estimate of the mixture density
# and the findings at a specific FDR level
Q = 0.1
plotFDR(fdr1, Q=Q, showfz=TRUE)

# Posterior distribution of the intercept
hist(fdr1$betasave[,1], 20)

# Compare actual versus estimated prior probabilities of being a signal
plot(wsuccess, fdr1$priorprob)

# Covariate effects
plot(X[,1], log(fdr1$priorprob/{1-fdr1$priorprob}), ylab='Logit of prior probability')
plot(X[,2], log(fdr1$priorprob/{1-fdr1$priorprob}), ylab='Logit of prior probability')

# Local FDR
plot(z, fdr1$localfdr, ylab='Local false-discovery rate')

# Extract findings at level FDR = Q
myfindings = which(fdr1$FDR <= Q)

## End(Not run)

FDRreg documentation built on May 2, 2019, 12:36 a.m.