hbfm.fit: Hierarchical Bayesian factor model MCMC sampler

In mnsekula/hbfm: Generating gene co-expression matrices from scRNA-seq data

Description

Function used to implement the MCMC sampler for the hierarchical Bayesian factor model (HBFM) defined in the "A sparse Bayesian factor model for the construction of gene co-expression networks from single-cell RNA sequencing count data" manuscript.

Usage

hbfm.fit(stoc.em.param, M = 4000, M.save = 1000, M.ll.seq = 10,
  H = 50, phi.scale = 0.25, par.samp = FALSE, seed = 123,
  verbose = FALSE)

Arguments

stoc.em.param	object of hbfm.par-class created by the stoc.em function
M	total number of MCMC iterations
M.save	number of iterations to be used for correlation estimation; the final M.save iterations from the M number of MCMC iterations will be saved for further analysis
M.ll.seq	intervals for calculating marginal log-likelihood before final M.save iterations; used to reduce computational time
H	number of lambda draws for marginal log-likelihood calculation
phi.scale	lognormal scale parameter used when drawing candidate values of phi in Metropolis-Hastings step; can be used to adjust acceptance rate of phi samples
par.samp	if TRUE, samples of each parameter in hbfm.fit for each MCMC iteration are saved; if FALSE only the calculated parameters of correlation and marginal log-likelihood are saved
seed	seed for random number generation
verbose	if TRUE, hbfm.fit status is displayed at every 100th iteration

Value

hbfm.fit-class object containing:

• Y: data.frame or matrix of gene expression counts

• Fac: number of factors considered in the model

• samples: samples of estimated parameters and/or calculated parameters from MCMC iterations

  • mll: calculated marginal log-likelihoods

  • corr: samples of gene-gene correlation matrix

  • beta: samples of beta parameters; included only when par.samp = TRUE

  • theta: samples of theta parameters; included only when par.samp = TRUE

  • alpha: samples of alpha parameters; included only when par.samp = TRUE

  • lambda: samples of lambda parameters; included only when par.samp = TRUE

  • phi: samples of phi parameters; included only when par.samp = TRUE

  • h1: samples of location hyperparameter of lognormal prior for phi; included only when par.samp = TRUE

  • h2: samples of scale hyperparameter of lognormal prior for phi; included only when par.samp = TRUE

Examples

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

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

## Run MCMC sampler with initial parameter values from stoc.em
fit.res1 <- hbfm.fit(fit1)

## End(Not run)

mnsekula/hbfm documentation built on June 29, 2020, 5:12 a.m.