iFad (version 2.0)

gibbs_sampling: Gibbs sampling for the inference of the inference of parameters in the sparse factor analysis model

Description

This function implements the collapsed Gibbs sampling algorithm for the inference of unknown parameters in the proposed sparse factor analysis model

Usage

gibbs_sampling(matrixY1, matrixY2, matrixL1, matrixL2, eta0, 
eta1, alpha_tau = 1, beta_tau = 0.01, tau_sig = 0, max_iter = 1e+05, 
thin = 10, alpha_sigma1 = 0.7, alpha_sigma2 = 0.7, 
beta_sigma1 = 0.3, beta_sigma2 = 0.3, file_name)

Arguments

matrixY1

The gene expression matrix, dim(Y1)=G1*J

matrixY2

The drug sensitivity matrix, dim(Y2)=G2*J

matrixL1

The linkage matrix representing prior knowledge about the sparsity pattern of matrixZ1

matrixL2

The linkage matrix representing prior knowledge about the sparsity pattern of matrixZ2

eta0

The probability of having true value of 1 for the entries in matrixZ with value 0 in matrixL

eta1

The probability of having true value of 0 for the entries in matrixZ with value 1 in matrixL

alpha_tau

The alpha parameter of Gamma distribution used for the simulation of noise, default value=1

beta_tau

The beta parameter of Gamma distribution used for the simulation of noise, default value=0.01

tau_sig

Pre-defined precision of each entry in the factor loadings matrixW1 and matrixW2, default value=0

max_iter

The number of iterations of the collaped Gibbs sampling algorithm, default=100000

thin

The number of iteration cycle for the record of Gibbs samples. For the convenience of storage, the result of the Gibbs sampling will be kept every other "thin" iterations to alliviate the auto-correlation

alpha_sigma1,alpha_sigma2

If tau_sig=0, the precision of each entry in the factor loading matrixW1 and matrixW2 is not pre-defined, but also treated in a bayesian way. The implemented algorithm will then put a Gamma p

beta_sigma1,beta_sigma2

The alpha parameter for the Gamma prior for matrixW1 and matrixW2 respectively

file_name

The name of the file to store the result of each thinned iteration of the Gibbs sampling

Value

• A ".RData" file with the following components:
• matrixZ_chain
  The updated matrixZ in each recorded iteration.A list of length 2, for matrixZ1 and matrixZ2 respectively. matrixZ_chain[[1]] and matrixZ2[[2]] are both matrices with dimension A*B, whereas A is the number of recorded iterations and B the length of matrixZ1/Z2. Each row of the matrix corresponds to the vectorized matrixZ1 or matrixZ2 in each iteration.
• matrixW_chain
  The updated matrixW in each recorded iteration, with format similar to matrixZ_chain.
• matrixX_chain
  The updated matrixX in each recorded iteration. An A*B matrix with each row corresponding to the vectorized matrixX in each recorded iteration.
• tau_g_chain
  The updated precision of each gene in each itertaion. A list with length 2, containing 2 matrices. Each row corresponds to the updated precision of the G1 or G2 genes in each recorded iteration.
• matrixPr_chain
  Posterior probability for each entry in matrixZ to take value of 1 in each recorded iteration. Also a list of length 2. Each element is a matrix of the same dimension with matrixZ_chain[[1]] or [[2]].
• label_chain
  A matrix to record the factor labels updated in each iteration. The label index is relative to the starting configuration in the Gibbs sampling process, not the true factor labels.

Examples

library(Rlab)
library(MASS)
library(coda)
library(ROCR)

data(matrixY1)
data(matrixY2)
data(matrixL1)
data(matrixL2)

gibbs_sampling(matrixY1, matrixY2, matrixL1, matrixL2, 
eta0=c(0.2,0.2), eta1=c(0.2,0.2), alpha_tau = 1, 
beta_tau = 0.01, tau_sig = 1, max_iter = 5, thin = 1, 
file_name="Demo_Gibbs_result.RData")