BFR.BE.EM: Bayesian factor regression with batch effect correction.

In AleAviP/BFR.BE: Bayesian factor regression with batch effect correction.

Description

Bayesian factor regression with batch effect correction.

Usage

BFR.BE.EM(x, v = NULL, b = NULL, q = 2, eps = 0.001, it = 100,
  epsM = 0.05, prior = "N.MOM", varianceBE = TRUE, init = NULL,
  hyper = NULL, seed = 4, scaling = FALSE, intercept = FALSE,
  varimax = FALSE)

Arguments

x	n x p matrix of p observations for n individuals
v	n x p_v matrix of known covariates.
b	n x p_b matrix of batch indicator.
q	latent cardinality.
eps	margin for convergence.
it	maximum number of iterations.
epsM	margin for convergence for M.
prior	Spike prior to use for the laodings: "N.MOM", "Normal" or "Flat".
varianceBE	TRUE if variance batch correction is modeled; FALSE if not.
init	list of initial parameters.
hyper	list of hyper parameters.
seed	random seed to use.
scaling	mean centering the data?
intercept	adds intercept to the Bayesian factor regression.
varimax	TRUE for initial varimax rotated loadings.

Value

A list with the optimised parameters.

Examples

n<-200
p<-250
q<-10
Pv = 1
Pb = 2

grid<-seq(-1,1,length.out=p)
M<-matrix(grid,ncol=q,nrow=p)
rate<-trunc(p/(q*2))
for(i in 2:q){
  M[,i]<-grid[c((i*rate):p,1:(i*rate-1))]
}

Z <- rmvnorm(n,numeric(q),diag(q))
Psi <- diag(c(rep(.5,p)),p)
V <- matrix(ncol = Pv, nrow = n, runif(n*Pv,0,3))
Theta = matrix(ncol=Pv,c(rep(-2,round(p/2)),rep(2,p-round(p/2))))
batch = round(runif(n,0,1),0)
B = matrix(ncol=Pb, c(batch,ifelse(batch==1,0,1)))
Beta = matrix(ncol=Pb, c(rep(2,p),rep(0,p)))
Tau_inv = cbind(diag(Psi),diag(Psi)*1.5)
Er = matrix(ncol=p, nrow=n)
for (i in 1:n){
  if(i%%10==0){print(i)}
  Er[i,] = mvrnorm(1, rep(0,p), diag(c(Tau_inv %*% B[i,]),p), tol = 1e-6, empirical = FALSE, EISPACK = FALSE)
}

X <- Z%*%t(M) + V%*%t(Theta) + B%*%t(Beta) + Er

result=BFR.BE.EM(x=X,v=V,b=B,q = 10)

AleAviP/BFR.BE documentation built on Feb. 22, 2020, 12:34 p.m.