# FDRreg-package: False discovery rate regression In FDRreg: False discovery rate regression

## 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.