BayesMP: MCMC for BayesMP
In Caleb-Huo/BayesMP: An R package perform BayesMP

Description Usage Arguments Details Value Author(s) Examples

View source: R/BayesMP.r

BayesMP

BayesMP(Z, gamma = NULL, updateGamma = TRUE, beta = 1/2, alpha = 1,
  mu0 = 0, sigma0 = 10, sigma = 1, trunc = 0, empMu = rep(0, ncol(Z)),
  empSD = rep(1, ncol(Z)), niter = 100, burnin = 50, silence = FALSE,
  logDotsPerLine = 50, fileName = "BayesMP_", writeY = TRUE,
  writePi = FALSE, writeDelta = FALSE, writeGamma = FALSE,
  writeHSall = TRUE)

`Z`	Z statistics. Z should be a p*n matrix.
`gamma`	initial gamma Estimated null proportation, by default, gamma will be estimated by the emperical null method.
`updateGamma`	If TRUE, will update gamma by MH method. If FALSE, will keep the gamma as constant.
`beta`	Non-informative prior: given a gene is DE, the prior probablity this gene is up-regulated, default=1/2.
`alpha`	Concentration parameter for DPs, default=1.
`mu0`	Mean parameter for base function, default=0.
`sigma0`	sqrt root of variance parameter for base function, default=10.
`sigma`	sqrt root of variance parameter for DP mixture component, default=1.
`trunc`	truncation parameter for base function (For both positive component and negative component), default=0.
`empMu`	a vector of mean parameter for the null component, default is 0. Alternatively, this vector can by estimated by the emperical null method.
`empSD`	a vector of sd parameter for the null component, default is 1. Alternatively, this vector can by estimated by the emperical null method.
`niter`	Number of iterations. Default 100, suggest to be 10,000
`burnin`	Number of burnin period. Default 50, suggest to be 500
`silence`	If FALSE (default), will print the MCMC progress in the console.
`logDotsPerLine`	Number of dots printed perline in the console when silence is FALSE.
`fileName`	Base fileName for saving fulll mcmc results.
`writeY`	If TRUE, will save all (niter) posterior samples of Y from MCMC.
`writePi`	If TRUE, will save all (niter) posterior samples of Pi from MCMC.
`writeDelta`	If TRUE, will save all (niter) posterior samples of Delta from MCMC.
`writeGamma`	If TRUE, will save all (niter) posterior samples of Gamma from MCMC.
`writeHSall`	If TRUE, will save the HSall matrix (niter - burnin). Each row of HSall represent a input gene (feature). If the number in the ith column and jth row equals m, it represents there are m posterior samples of Y for feature j that are DE in at least i studies. This matrix will be the input matrix to calculate the Bayesian FDR. For example, for each column i, after normalized (divided) by total number effective samples (niter - burnin), it is the Bayesian belief that the genes are significiant in at least i studies. The Bayesian FDR can be calculated by the BayesianFDR function.

implementation for BayesMP, MCMC part. This is a full Bayesian model, and the alternative distribution is modeled via dirichlet process.

The MCMC object for the last iteration.

Zhiguang Huo <zhuo@ufl.edu>

set.seed(15213)
G <- 2000
K <- 10
alpha <- 200
X0 <- matrix(rnorm(G * K), G, K)
Xplus <- matrix(rnorm(G * K, 2), G, K)
Xminus <- matrix(rnorm(G * K, -2), G, K)
piall <- rbeta(G, alpha/G, 1)
delta <- rbeta(G, 1/2, 1/2)
p0 <- 1 - piall
p1 <- piall * delta
p2 <- piall * (1 - delta)
Y <- replicate(K, apply(cbind(p0, p1, p2),1,function(x) sample(c(0,1,-1),1,prob = x)))
Z <- X0 * (Y == 0) + Xplus * (Y == 1) + Xminus * (Y == -1)
system.time(BayesMP(Z,writeHSall=F))