variance_prior: Estimate the scale and df for a Inverse Chisquare...

In const-ae/glmGamPoi: Fit a Gamma-Poisson Generalized Linear Model

Estimate the scale and df for a Inverse Chisquare distribution that generate the true gene variances

Description

This function implements Smyth's 2004 variance shrinkage. It also supports covariates that are fitted to log(s2) with natural splines. This is based on the 2012 Lund et al. quasi-likelihood paper.

Usage

variance_prior(s2, df, covariate = NULL, abundance_trend = NULL)

Arguments

s2	vector of observed variances. Must not contain 0's.
df	vector or single number with the degrees of freedom
covariate	a vector with the same length as s2. covariate is used to regress out the trend in s2. If covariate = NULL, it is ignored.
abundance_trend	logical that decides if the additional abundance trend is fit to the data. If NULL the abundance trend is fitted if there are more than 10 observations and the covariate is not NULL. Default: NULL

Value

a list with three elements:

variance0

estimate of the scale of the inverse Chisquared distribution. If covariate is NULL a single number, otherwise a vector of length(covariate)

df0

estimate of the degrees of freedom of the inverse Chisquared distribution

var_pos

the shrunken variance estimates: a combination of s2 and variance0

See Also

limma::squeezeVar()

const-ae/glmGamPoi documentation built on July 25, 2024, 10:22 p.m.