ASSIGN (version 1.8.0)

assign.mcmc: The Gibbs sampling algorithm to approximate the joint distribution of the model parameters

Description

The assign.mcmc function uses a Bayesian sparse factor analysis model to estimate the adaptive baseline/background, adaptive pathway signature, and pathway activation status of individual test (disease) samples.

Usage

assign.mcmc(Y, Bg, X, Delta_prior_p, iter=2000, adaptive_B=TRUE, adaptive_S=FALSE,
mixture_beta=TRUE, sigma_sZero = 0.01, sigma_sNonZero = 1, p_beta = 0.01, 
sigma_bZero = 0.01, sigma_bNonZero = 1, alpha_tau = 1, beta_tau = 0.01, 
Bg_zeroPrior=TRUE, S_zeroPrior=TRUE, ECM = FALSE)

Arguments

Y

The G x J matrix of genomic measures (i.g., gene expession) of test samples. Y is the testData_sub variable returned from the data.process function. Genes/probes present in at least one pathway signature are retained.

Bg

The G x 1 vector of genomic measures of the baseline/background (B). Bg is the B_vector variable returned from the data.process function. Bg is the starting value of baseline/background level in the MCMC chain.

X

The G x K matrix of genomic measures of the signature. X is the S_matrix variable returned from the data.process function. X is the starting value of pathway signatures in the MCMC chain.

Delta_prior_p

The G x K matrix of prior probability of a gene being "significant" in its associated pathway. Delta_prior_p is the Pi_matrix variable returned from the data.process function.

adaptive_B

Logicals. If TRUE, the model adapts the baseline/background (B) of genomic measures for the test samples. The default is TRUE.

adaptive_S

Logicals. If TRUE, the model adapts the signatures (S) of genomic measures for the test samples. The default is FALSE.

iter

The number of iterations in the MCMC. The default is 2000.

sigma_sZero

Each element of the signature matrix (S) is modeled by a spike-and-slab mixuture distribution. Sigma_sZero is the variance of the spike normal distribution. The default is 0.01.

sigma_sNonZero

Each element of the signature matrix (S) is modeled by a spike-and-slab mixuture distribution. Sigma_sNonZero is the variance of the slab normal distribution. The default is 1.

p_beta

p_beta is the prior probability of a pathway being activated in individual test samples. The default is 0.01.

sigma_bZero

Each element of the pathway activation matrix (A) is modeled by a spike-and-slab mixuture distribution. sigma_bZero is the variance of the spike normal distribution. The default is 0.01.

sigma_bNonZero

Each element of the pathway activation matrix (A) is modeled by a spike-and-slab mixuture distribution. sigma_bNonZero is the variance of the slab normal distribution. The default is 1.

alpha_tau

The shape parameter of the precision (inverse of the variance) of a gene. The default is 1.

beta_tau

The rate parameter of the precision (inverse of the variance) of a gene. The default is 0.01.

Bg_zeroPrior

Logicals. If TRUE, the prior distritribution of baseline/background level follows a normal distribution with mean zero. The default is TRUE.

mixture_beta

Logicals. If TRUE, elements of the pathway activation matrix are modeled by a spike-and-slab mixuture distribution. The default is TRUE.

S_zeroPrior

Logicals. If TRUE, the prior distritribution of signature follows a normal distribution with mean zero. The default is TRUE.

ECM

Logicals. If TRUE, ECM algorithm, rather than Gibbs sampling, is applied to approximate the model parameters. The default is FALSE.

Value

beta_mcmc

The iter x K x J array of the pathway activation level estimated in every iteration of MCMC.

tau2_mcmc

The iter x G matrix of the precision of genes estimated in every iteration of MCMC

gamma_mcmc

The iter x K x J array of probability of pathway being activated estimated in every iteration of MCMC.

kappa_mcmc

The iter x K x J array of pathway activation level (adjusted beta scaling between 0 and 1) estimated in every iteration of MCMC.)

S_mcmc

The iter x G x K array of signature estimated in every iteration of MCMC.

Delta_mcmc

The iter x G x K array of binary indicator of a gene being significant estimated in every iteration of MCMC.

Details

The assign.mcmc function can be set as following major modes. The combination of logical values of adaptive_B, adaptive_S and mixture_beta can form different modes.

Mode A: adaptive_B = FALSE, adaptive_S = FALSE, mixture_beta = FALSE. This is a regression mode without adaptation of baseline/background, signature, and no shrinkage of the pathway activation level.

Mode B: adaptive_B = TRUE, adaptive_S = FALSE, mixture_beta = FALSE. This is a regression mode with adaptation of baseline/background, but without signature, and with no shrinkage of the pathway activation level.

Mode C: adaptive_B = TRUE, adaptive_S = FALSE, mixture_beta = TRUE. This is a regression mode with adaptation of baseline/background, but without signature, and with shrinkage of the pathway activation level when it is not significantly activated.

Mode D: adaptive_B = TRUE, adaptive_S = TRUE, mixture_beta = TRUE. This is a Bayesian factor analysis mode with adaptation of baseline/background, adaptation signature, and with shrinkage of the pathway activation level.

Examples

mcmc.chain <- assign.mcmc(Y=processed.data$testData_sub, Bg = processed.data$B_vector, 
X=processed.data$S_matrix, Delta_prior_p = processed.data$Pi_matrix, iter = 20, 
adaptive_B=TRUE, adaptive_S=FALSE, mixture_beta=TRUE)