ebdbNet (version 1.1)

ebdbNet-package: Empirical Bayes Dynamic Bayesian Network (EBDBN) Inference

Description

This package is used to infer the adjacency matrix of a network from time course data using an empirical Bayes estimation procedure based on Dynamic Bayesian Networks.

Details

ll{ Package: ebdbNet Type: Package Version: 1.1 Date: 2010-01-20 License: GPL (version 3 or later) LazyLoad: yes } Posterior distributions (mean and variance) of network parameters are estimated using time-course data based on a linear feedback state space model that allows for a set of hidden states to be incorporated. The algorithm is composed of three principal parts: choice of hidden state dimension (see hankel), estimation of hidden states via the Kalman filter and smoother, and calculation of posterior distributions based on the empirical Bayes estimation of hyperparameters in a hierarchical Bayesian framework (see ebdbn).

References

Andrea Rau, Florence Jaffrezic, Jean-Louis Foulley, and R. W. Doerge (2010). An Empirical Bayesian Method for Estimating Biological Networks from Temporal Microarray Data. Statistical Applications in Genetics and Molecular Biology 9. Article 9.

Examples

library(ebdbNet)
library(GeneNet)  ## Load GeneNet package to use T-cell activation data
tmp <- runif(1) ## Initialize random number generator
set.seed(4568818) ## Set seed

## Load T-cell activation data
data(tcell) 
tc44 <- combine.longitudinal(tcell.10, tcell.34)

## Put data into correct format for algorithm
## (List, with one matrix per replicate (P rows and T columns)
tcell.dat <- dataFormat(tc44)

## Use only subset of T-cell data for faster example
R <- 5 ## 5 replicates
T <- 10 ## 10 time points
P <- 15 ## 15 genes
tcell.sub.dat <- vector("list", R)
for(r in 1:R) {
	tcell.sub.dat[[r]] <- tcell.dat[[r]][1:P, 1:T]
}

####################################################
# Run EBDBN (no hidden states) with feedback loops
####################################################
## Choose alternative value of K using hankel if hidden states to be estimated
## K <- hankel(tcell.sub.dat, lag = 1)$dim

## Run algorithm
net <- ebdbn(input = "feedback", tcell.sub.dat, K = 0, conv.1 = 0.10,
	conv.2 = 0.01, conv.3 = 0.01)

## Use z-score significance level of 99%
z <- zCutoff(net$DPost, net$DvarPost)$z99