fblr {scrime}R Documentation

Full Bayesian Logic Regression for SNP Data

Description

Performs full Bayesian logic regression for Single Nucleotide Polymorphism (SNP) data as described in Fritsch and Ickstadt (2007).

fblr.weight allows to incorporate prior pathway information by restricting search for interactions to specific groups of SNPs and/or giving them different weights. fblr.weight is only implemented for an interaction level of 2.

Usage

fblr(y, bin, niter, thin = 5, nburn = 10000, int.level = 2, kmax = 10, 
  geo = 1, delta1 = 0.001, delta2 = 0.1, predict = FALSE, 
  file = "fblr_mcmc.txt")

fblr.weight(y, bin, niter, thin = 5, nburn = 10000, kmax = 10, geo = 1, 
  delta1 = 0.001, delta2 = 0.1, predict = FALSE, group = NULL, 
  weight = NULL, file = "fblr_mcmc.txt")

Arguments

y binary vector indicating case-control status.
bin binary matrix with number of rows equal to length(y). Usually the result of applying snp2bin to a matrix of SNP data.
niter number of MCMC iterations after burn-in.
thin after burn-in only every thinth iteration is kept.
nburn number of burn-in iterations.
int.level maximum number of binaries allowed in a logic predictor. Is fixed to 2 for fblr.weight.
kmax maximum number of logic predictors allowed in the model.
geo geometric penalty parameter for the number of binaries in a predictor. Value between 0 and 1. Default is 1, meaning no penalty.
delta1 shape parameter for hierarchical gamma prior on precision of regression parameters.
delta2 rate parameter for hierarchical gamma prior on precision of regression parameters.
predict should predicted case probabilities be returned?
file character string naming a file to write the MCMC output to. If fblr is called again, the file is overwritten.
group list containing vectors of indices of binaries that are allowed to interact. Groups may be overlapping, but every binary has to be in at least one group. Groups have to contain at least two binaries. Defaults to NULL, meaning that all interactions are allowed.
weight vector of length ncol(bin) containing different relative prior weights for binaries. Defaults to NULL, meaning equal weight for all binaries.

Details

The MCMC output in file can be analysed using the function analyse.models. In the help of this function it is also described how the models are stored in file.

Value

accept acceptance rate of MCMC algorithm.
pred vector of predicted case probabilities. Only given if predict = TRUE.

Author(s)

Arno Fritsch, arno.fritsch@uni-dortmund.de

References

Fritsch, A. and Ickstadt, K. (2007). Comparing logic regression based methods for identifying SNP interactions. In Bioinformatics in Research and Development, Hochreiter, S. and Wagner, R. (Eds.), Springer, Berlin.

See Also

analyse.models,predictFBLR

Examples

## Not run: 
# SNP dataset with 500 persons and 20 SNPs each,
# a two-SNP interaction influences the case probability
snp <- matrix(rbinom(500*20,2,0.3),ncol=20)
bin <- snp2bin(snp)
int <- apply(bin,1,function(x) (x[1] == 1 & x[3] == 0)*1)
case.prob <- exp(-0.5+log(5)*int)/(1+exp(-0.5+log(5)*int))
y <- rbinom(nrow(snp),1,prob=case.prob)

# normally more iterations should be used
fblr(y, bin, niter=1000, nburn=0)
analyse.models("fblr_mcmc.txt")

# Prior information: SNPs 1-10 belong to genes in one pathway, 
# SNPs 8-20 to another. Only interactions within a pathway are 
# considered and the first pathway is deemed to be twice as 
# important than the second.
fblr.weight(y,bin,niter=1000, nburn=0, group=list(1:20, 15:40), 
  weight=c(rep(2,20),rep(1,20)))
analyse.models("fblr_mcmc.txt")

## End(Not run)

[Package scrime version 1.1.2 Index]