mmcmcBayes (version 0.1.0)

asgn_func: ASGN Model Parameter Estimation using MCMC

Description

Estimates parameters (alpha, mu, sigma2) for the ASGN model using Markov Chain Monte Carlo (MCMC) methods. This function can be used for both cancer and normal groups in the multi-stage DMR detection framework, and also as a standalone function for fitting skewed and potentially bimodal data.

Usage

asgn_func(
  data,
  priors = NULL,
  mcmc = list(nburn = 5000, niter = 10000, thin = 1),
  seed = NULL,
  return_mcmc = FALSE,
  return_summary = FALSE
)

Value

A list that may contain the following elements:

• posteriors: Vector of posterior means for alpha, mu, and sigma2 (always present)

• mcmc_samples: List containing full MCMC chains for alpha, mu, and sigma2 (only if return_mcmc = TRUE)

• summary: Data.frame containing parameter estimates with 95% credible intervals (only if return_summary = TRUE)

Arguments

data

A matrix or vector containing the mean methylation levels (M-values)

priors

A list of prior parameters for alpha, mu, and sigma2. If NULL, weakly informative priors are automatically generated from the data.

mcmc

A list of MCMC parameters:

• nburn: Number of burn-in iterations (default: 5000)

• niter: Number of sampling iterations (default: 10000)

• thin: Thinning interval (default: 1)

seed

Random seed for reproducibility. If NULL, no seed is set.

return_mcmc

Logical indicating whether to return full MCMC samples for diagnostic purposes (default: FALSE)

return_summary

Logical indicating whether to return a summary data.frame with parameter estimates and 95% credible intervals (default: FALSE)

Author

Zhexuan Yang, Duchwan Ryu, and Feng Luan

Details

This function implements a Metropolis-Hastings within Gibbs sampler to estimate the parameters of the ASGN distribution, which can model skewed and potentially bimodal data. The algorithm updates alpha and mu using Metropolis-Hastings steps and sigma2 using inverse Gamma sampling.

The ASGN (Alpha-skewed generalized normal) distribution is particularly useful for modeling methylation data that often exhibits skewness and bimodal issue.

See Also

mmcmcBayes for the main DMR detection function, traceplot_asgn for MCMC diagnostic plots

Examples

# \donttest{
# Generate sample data
set.seed(2021)
dt <- rgamma(1000, shape = 2, rate = 1)
dt <- as.matrix(dt, ncol=1)

result <- asgn_func(dt, return_mcmc = TRUE, return_summary = TRUE)
print(result$summary)
# }