pcapM: A wrapper for pcaMethods function implementations

In HGray384/pcaNet: Probabilistic principal components analysis - covariance estimation and network reconstruction

Description

Implements the equivalent of pca. This function preprocesses the data as specified by the user, then calls ppcapM or bpcapM, and finally handles this output to return a list. One element of the output is a pcaRes object.

Usage

pcapM(
  myMat,
  nPcs = 2,
  method = "ppca",
  seed = NA,
  threshold = 1e-04,
  maxIterations = 1000,
  center = TRUE,
  scale = c("none"),
  loglike = TRUE,
  verbose = TRUE
)

Arguments

myMat	matrix – Data matrix with variables in columns and observations in rows. The data may contain missing values, denoted as NA.
nPcs	numeric – Number of components used for re-estimation. Choosing few components may decrease the estimation precision.
method	c("ppca", "bpca") – frequentist or Bayesian estimation of model parameters.
seed	numeric – the random number seed used, useful to specify when comparing algorithms.
threshold	numeric – Convergence threshold. If the increase in precision of an update falls below this then the algorithm is stopped.
maxIterations	numeric – Maximum number of estimation steps.
center	logical – should the data be centered?
scale	c("none", "pareto", "vector", "uv") – which method of scaling should be used? See pca. Defaults to "none".
loglike	logical – should the log-likelihood of the estimated parameters be returned? See Details.
verbose	logical – verbose intermediary algorithm output.

Details

See ppcapM and bpcapM for the algorithm specifics. loglike indicates whether log-likelihood values for the resulting estimates should be computed. This can be useful to compare different algorithms.

Value

A list of 5 or 7 elements, depending on the value of loglike:

W

matrix – the estimated loadings.

sigmaSq

numeric – the estimated isotropic variance.

Sigma

matrix – the estimated covariance matrix.

m

numeric – the estimated mean vector.

logLikeObs

numeric – the log-likelihood value of the observed data given the estimated parameters.

logLikeImp

numeric – the log-likelihood value of the imputed data given the estimated parameters.

pcaMethodsRes

class – see pcaRes.

Examples

# simulate a dataset from a zero mean factor model X = Wz + epsilon
# start off by generating a random binary connectivity matrix
n.factors <- 5
n.genes <- 200
# with dense connectivity
# set.seed(20)
conn.mat <- matrix(rbinom(n = n.genes*n.factors,
                          size = 1, prob = 0.7), c(n.genes, n.factors))

# now generate a loadings matrix from this connectivity
loading.gen <- function(x){
  ifelse(x==0, 0, rnorm(1, 0, 1))
}

W <- apply(conn.mat, c(1, 2), loading.gen)

# generate factor matrix
n.samples <- 100
z <- replicate(n.samples, rnorm(n.factors, 0, 1))

# generate a noise matrix
sigma.sq <- 0.1
epsilon <- replicate(n.samples, rnorm(n.genes, 0, sqrt(sigma.sq)))

# by the ppca equations this gives us the data matrix
X <- W%*%z + epsilon
WWt <- tcrossprod(W)
Sigma <- WWt + diag(sigma.sq, n.genes)

# select 10% of entries to make missing values
missFrac <- 0.1
inds <- sample(x = 1:length(X),
               size = ceiling(length(X)*missFrac),
               replace = FALSE)

# replace them with NAs in the dataset
missing.dataset <- X
missing.dataset[inds] <- NA

# run ppca
ppm <- pcapM(t(missing.dataset), nPcs=5, method="bpca", seed=2009, 
maxIterations=1000, center=TRUE, loglike=TRUE, verbose=TRUE)

HGray384/pcaNet documentation built on Nov. 14, 2020, 11:11 a.m.