zinbInitialize: Initialize the parameters of a ZINB regression model

In zinbwave: Zero-Inflated Negative Binomial Model for RNA-Seq Data

Description

The initialization performs quick optimization of the parameters with several simplifying assumptions compared to the true model: non-zero counts are models as log-Gaussian, zeros are modeled as dropouts. The dispersion parameter is not modified.

Usage

zinbInitialize(
  m,
  Y,
  nb.repeat = 2,
  it.max = 100,
  BPPARAM = BiocParallel::bpparam()
)

Arguments

m	The model of class ZinbModel
Y	The matrix of counts.
nb.repeat	Number of iterations for the estimation of beta_mu and gamma_mu.
it.max	Maximum number of iterations in softImpute.
BPPARAM	object of class bpparamClass that specifies the back-end to be used for computations. See bpparam for details.

Value

An object of class ZinbModel similar to the one given as argument with modified parameters alpha_mu, alpha_pi, beta_mu, beta_pi, gamma_mu, gamma_pi, W.

Examples

Y <- matrix(rpois(60, lambda=2), 6, 10)
bio <- gl(2, 3)
time <- rnorm(6)
gc <- rnorm(10)
m <- zinbModel(Y, X=model.matrix(~bio + time), V=model.matrix(~gc),
             which_X_pi=1L, which_V_mu=1L, K=1)
m <- zinbInitialize(m, Y, BPPARAM=BiocParallel::SerialParam())

zinbwave documentation built on Nov. 8, 2020, 8:11 p.m.