micore: Run Microbiome Covariance Regression

In kevinmcgregor/micore: Microbiome Covariance Regression

Description

Run Microbiome Covariance Regression

Usage

micore(
  counts,
  X,
  C0 = NULL,
  Psi0 = NULL,
  Gamma0 = NULL,
  nuPsi = NULL,
  nuGamma = NULL,
  n.chain = 4,
  n.cores = n.chain,
  target.accept.rate = 0.23,
  n.samp = 4000,
  n.burn = 4000,
  adapt.control = NULL,
  save.eta.cov = FALSE,
  verbose = FALSE
)

Arguments

counts	Matrix containing OTU counts. Rows are samples, columns are OTUs. The last column is considered the reference OTU for the additive log-ratio transformation and is not included in the covariance matrix.
X	Design matrix contating covariates of interest. Rows are samples, columns are covariates. Can include intercept column.
C0	Prior mean for "C" parameter. If null, then C0 is set automatically.
Psi0	Prior inverse-Wishart matrix hyperparameter for "Psi" parameter. If null, then Psi0 is set automatically.
Gamma0	Prior inverse-Wishart matrix hyperparameter for "Gamma" parameter. If null, then Gamma0 is set automatically.
nuPsi	Prior inverse-Wishart degrees of freedom for "Psi" parameter. If null, then nuPsi is set automatically.
nuGamma	Prior inverse-Wishart degrees of freedom for "Gamma" parameter. If null, then nuGamma is set automatically.
n.chain	Number of independent MCMC chains to run in parallel
n.cores	Number of cores to use to parallelize the MCMC chains
target.accept.rate	Target acceptance rate for adaptive Metropolis sampler.
n.samp	Number of MCMC samples
n.burn	Number of burn-in samples
adapt.control	List contatining tuning parameters for adaptive MCMC part. See details section below.
save.eta.cov	Logical. Save the proposal distribution variance from the Metropolis sampler? Caution, this will be large.
verbose	Logical. Output progress?

Details

This function allows running the Microbiome Covariance Regession method on multiple chains in parallel. The user can specify values for the hyperparameters C0, Psi0, Gamma0, nuPsi, and nuGamma. If they are not specified, then they will be set automatically.

A list of arguments passed to the adaptive Metropolis sampler for eta can be input using the adapt.control argument. These can be modified if the eta parameters do not converge; the default values are not always optimal. The parameters (and defaults) that can be edited are:

• init: (default 0.1) The initial stepsize for the adaptive Metropolis parameters.

• a: (default 0.5) The adaptation rate. A higher value means the adaptations will vanish more quickly with the number of MCMC steps.

• sigma.zero: (default 1) The initial value of the adaptive Metropolis variance scaling parameter.

Value

An object of class micore containing one list element for each MCMC chain. Each list element contains the following attributes:

• eta: An array containing the MCMC samples for the eta parameter. First dimension indexes the MCMC samples.

• Psi: An array containing the MCMC samples for the Psi parameter. First dimension indexes the MCMC samples.

• A: An array containing the MCMC samples for the A parameter. First dimension indexes the MCMC samples.

• B: An array containing the MCMC samples for the B parameter. First dimension indexes the MCMC samples.

• gamma: An array containing the MCMC samples for the gamma_i parameters. First dimension indexes the MCMC samples.

• Gamma: An array containing the MCMC samples for the Gamma (matrix) parameter. First dimension indexes the MCMC samples.

• eta.accepted: A matrix indicating which eta samples were accepted (1) and which were rejected (0). Rows are MCMC iterations, columns are subjects.

• sigma.zero: A matrix containing values of the Metropolis variance scaling parameter for all subjects over all MCMC samples.

• acc.probs: A matrix containing the Metropolis acceptance probabilities for all subjects over all MCMC samples.

• counts: The OTU count matrix.

• X: The model matrix.

Examples

n <- 50
p <- 5
X <- cbind(1, rnorm(n))
counts <- matrix(0, n, p+1)
for (i in 1:n) {
  counts[i,] <- rmultinom(1, size=100, prob=rep(1,p+1))
}

library(micore)
mc.fit <- micore(counts, X, n.samp=100, n.burn=100, n.chain=1)

kevinmcgregor/micore documentation built on June 9, 2021, 10:29 p.m.