slfm {slfm}

slfm

Description

Bayesian Sparse Latent Factor Model (SLFM) designed for the analysis of coherent patterns in gene expression data matrices. Details about the methodology being applied here can be found in Duarte and Mayrink (2015) and Duarte and Mayrink (2019).

Usage

slfm(
  x,
  a = 2.1,
  b = 1.1,
  gamma_a = 1,
  gamma_b = 1,
  omega_0 = 0.01,
  omega_1 = 10,
  sample = 1000,
  burnin = round(0.25 * sample),
  lag = 1,
  degenerate = FALSE
)

Arguments

x

matrix with the pre-processed data.

a

positive shape parameter of the Inverse Gamma prior distribution (default = 2.1).

b

positive scale parameter of the Inverse Gamma prior distribution (default = 1.1).

gamma_a

positive 1st shape parameter of the Beta prior distribution (default = 1).

gamma_b

positive 2nd shape parameter of the Beta prior distribution (default = 1).

omega_0

prior variance of the spike mixture component (default = 0.01).

omega_1

prior variance of the slab mixture component (default = 10).

sample

sample size to be considered for inference after the burn in period (default = 1000).

burnin

size of the burn in period in the MCMC algorithm (default = sample/4).

lag

lag to build the chains based on spaced draws from the Gibbs sampler (defaul = 1).

degenerate

logical argument (default = FALSE) indicating whether to use the degenerate version of the mixture prior for the factor loadings.

Value

x: data matrix.

q_star: matrix of MCMC chains for q_star parameter.

alpha: summary table of MCMC chains for alpha parameter.

lambda: summary table of MCMC chains for lambda parameter.

sigma: summary table of MCMC chains for sigma parameter.

classification: classification of each alpha ('present', 'marginal', 'absent')

References

DOI:10.18637/jss.v090.i09 (Duarte and Mayrink; 2019)

DOI:10.1007/978-3-319-12454-4_15 (Duarte and Mayrink; 2015)

See Also

process_matrix, plot_matrix

Examples

mat <- matrix(rnorm(2000), nrow = 20)
slfm(mat, sample = 1000)