sBFAC: The sparse Bayesian FAC model.

In syspremed/PhenMAP: sBFAC - a sparse Bayesian FAC model for clustering expression data, finding associated phenotypes and molecular features.

The sparse Bayesian FAC model.

Description

A statistical tool for clustering expression data, finding associated phenotypes and molecular features.

Usage

sBFAC(Y, X, fmin = 1, fmax = 2, gmin = 1, gmax = 2, w0 = 1, g0 = nrow(Y), mcmc = 20000)

Arguments

Y	Expression data in samples by features format.
X	Design matrix from model.matrix function which defines how covariates are included into sBFAC.
fmin	Minimum number of latent variables to be considered. Default is 1.
fmax	Maximum number of latent variables to be considered. Default is 2. fmax should be greater or equal to fmin.
gmin	Minimum number of subtypes to be considered. Default is 1.
gmax	Maximum number of subtypes to be considered. Default is 2.
w0	Controls sparseness in features, minimum value is 0.01 for non-informative priors.
g0	Decreasing this regularize regression coefficients, i.e., Very large value of g0 provides non-informative prior == Classical estimates.
mcmc	Number of MCMC samples. Default is 20000 but can be increased to improve convergence.

Details

The function first select the optimal number of latent variables and fit sBFAC model via MCMC to identify subtypes and their associated phenotypes and molecular features.

Value

u.store	This is an array of q x N x mcmc containing scores for the q latent variables on N obaservations over mcmc samples.
w.store	This is an array of p x q x mcmc containing loadings for the p genes/features on q latent variables over mcmc samples.
b.store	This is an array of q x L x mcmc containing regression coefficients of L covariates on q latent variables over mcmc samples.
s.store	This is an matrix of p x mcmc containing residual variablity of each feature.
s2.store	This is an matrix of q x mcmc containing residual variablity of each latent variable.
MBClustering	Model based clustering of q latent variables.
KMeansClustering	KMeans clustering of q latent variables.

Author(s)

Gift Nyamundanda, Katie Eason, Pawan Poudel and Anguraj Sadanandam.

References

Nyamundanda et al (2016). A next generation tool enables identification of functional cancer subtypes with associated biological phenotypes.

Examples

## help(sBFAC)
## Expression and phenotype data are required for this function 
data(Y)
data(Cv)
## Create design matrix for the phenotye data
X<-model.matrix(~Etoposide+Fascaplysin+Bortezomib+Geldanamycin,Cv)  
## Run the function
sBFAC(Y,X,fmin=1,fmax=3,gmin=1,gmax=7,w0=1,g0=1000,mcmc=10000)

syspremed/PhenMAP documentation built on April 2, 2022, 3:12 p.m.