stoc.em: Stochastic EM algorithm for initial parameter estimation
In mnsekula/hbfm: Generating gene co-expression matrices from scRNA-seq data

Description Usage Arguments Details Value Examples

View source: R/hbfm_functions.R

Function used to implement the stochastic EM algorithm defined in the "A sparse Bayesian factor model for the construction of gene co-expression networks from single-cell RNA sequencing count data" manuscript.

1 2	stoc.em(Y, Fac, M.stoc = 2000, M.int = 100, M.eval = 200, M.ll.seq = 10, H = 50, seed = 123, verbose = FALSE)

`Y`	data.frame or matrix of gene expression counts where the rows correspond to genes and columns correspond to cells; Y must contain integers and have row names
`Fac`	number of factors to consider in the model; only a single number is accepted
`M.stoc`	total number of stochastic EM iterations
`M.int`	initial number of MCMC draws before maximization
`M.eval`	number of iterations to be used for parameter estimation; the final `M.eval` iterations from the `M.stoc` number of stochastic EM iterations will be considered
`M.ll.seq`	intervals for calculating marginal log-likelihood before final `M.eval` iterations; used to reduce computational time
`H`	number of lambda draws for marginal log-likelihood calculation
`seed`	seed for random number generation
`verbose`	if TRUE, `stoc.em` status is displayed at every 100th iteration

This algorithm should be used before hbfm.fit as it generates initial parameter values for use in hbfm.fit.

hbfm.par-class object containing:

Y: data.frame or matrix of gene expression counts
Fac: number of factors considered in the model
beta: initial beta parameter vector for input into hbfm.fit
theta: initial theta parameter vector for input into hbfm.fit
alpha: initial alpha parameter matrix for input into hbfm.fit
lambda: initial lambda parameter matrix for input into hbfm.fit
phi: initial phi parameter vector for input into hbfm.fit
h1: location hyperparameter of lognormal prior for phi to input into hbfm.fit
h2: scale hyperparameter of lognormal prior for phi to input into hbfm.fit
mll: calculated marginal log-likelihoods at given iterations; used for convergence diagnostics

## Not run: 
## Load dataset
data(gene.mat)

## Run stochastic EM first
## Consider F=5 factors
fit1 <- stoc.em(Y=gene.mat, Fac = 5)


## End(Not run)