melissa_gibbs: Gibbs sampling algorithm for Melissa model

In andreaskapou/Melissa: Bayesian clustering and imputationa of single cell methylomes

Description

melissa_gibbs implements the Gibbs sampling algorithm for performing clustering of single cells based on their DNA methylation profiles, where the observation model is the Bernoulli distributed Probit Regression likelihood. NOTE: that Gibbs sampling is really slow and we recommend using the VB implementation: melissa.

Usage

melissa_gibbs(
  X,
  K = 2,
  pi_k = rep(1/K, K),
  w = NULL,
  basis = NULL,
  w_0_mean = NULL,
  w_0_cov = NULL,
  dir_a = rep(1, K),
  lambda = 1/2,
  gibbs_nsim = 1000,
  gibbs_burn_in = 200,
  inner_gibbs = FALSE,
  gibbs_inner_nsim = 50,
  is_parallel = TRUE,
  no_cores = NULL,
  is_verbose = FALSE
)

Arguments

X	A list of length I, where I are the total number of cells. Each element of the list contains another list of length N, where N is the total number of genomic regions. Each element of the inner list is an L x 2 matrix of observations, where 1st column contains the locations and the 2nd column contains the methylation level of the corresponding CpGs.
K	Integer denoting the number of clusters K.
pi_k	Vector of length K, denoting the mixing proportions.
w	A N x M x K array, where each column contains the basis function coefficients for the corresponding cluster.
basis	A 'basis' object. E.g. see create_rbf_object from BPRMeth package
w_0_mean	The prior mean hyperparameter for w
w_0_cov	The prior covariance hyperparameter for w
dir_a	The Dirichlet concentration parameter, prior over pi_k
lambda	The complexity penalty coefficient for penalized regression.
gibbs_nsim	Argument giving the number of simulations of the Gibbs sampler.
gibbs_burn_in	Argument giving the burn in period of the Gibbs sampler.
inner_gibbs	Logical, indicating if we should perform Gibbs sampling to sample from the augmented BPR model.
gibbs_inner_nsim	Number of inner Gibbs simulations.
is_parallel	Logical, indicating if code should be run in parallel.
no_cores	Number of cores to be used, default is max_no_cores - 1.
is_verbose	Logical, print results during EM iterations

Value

An object of class melissa_gibbs.

Author(s)

C.A.Kapourani [e-mail]

See Also

melissa, create_melissa_data_obj, partition_dataset, filter_regions

Examples

# Example of running Melissa Gibbs on synthetic data

# Create RBF basis object with 4 RBFs
basis_obj <- BPRMeth::create_rbf_object(M = 4)

set.seed(15)
# Run Melissa Gibbs
melissa_obj <- melissa_gibbs(X = melissa_synth_dt$met, K = 2, basis = basis_obj,
   gibbs_nsim = 10, gibbs_burn_in = 5, is_parallel = FALSE, is_verbose = FALSE)

# Extract mixing proportions
print(melissa_obj$pi_k)

andreaskapou/Melissa documentation built on June 12, 2020, 5:54 p.m.