adjustedProfileLik: Adjusted Profile Likelihood for the Negative Binomial...

In edgeR: Empirical Analysis of Digital Gene Expression Data in R

Description

Compute adjusted profile log-likelihoods for the dispersion parameters of genewise negative binomial glms.

Usage

adjustedProfileLik(dispersion, y, design, offset, weights=NULL, adjust=TRUE, 
            start=NULL, get.coef=FALSE)

Arguments

dispersion	numeric scalar or vector of dispersions.
y	numeric matrix of counts.
design	numeric matrix giving the design matrix.
offset	numeric matrix of same size as y giving offsets for the log-linear models. Can be a scalor or a vector of length ncol(y), in which case it is expanded out to a matrix.
weights	optional numeric matrix giving observation weights.
adjust	logical, if TRUE then Cox-Reid adjustment is made to the log-likelihood, if FALSE then the log-likelihood is returned without adjustment.
start	numeric matrix of starting values for the GLM coefficients, to be passed to glmFit.
get.coef	logical, specifying whether fitted GLM coefficients should be returned.

Details

For each row of data, compute the adjusted profile log-likelihood for the dispersion parameter of the negative binomial glm. The adjusted profile likelihood is described by McCarthy et al (2012) and is based on the method of Cox and Reid (1987).

The adjusted profile likelihood is an approximation to the log-likelihood function, conditional on the estimated values of the coefficients in the NB log-linear models. The conditional likelihood approach is a technique for adjusting the likelihood function to allow for the fact that nuisance parameters have to be estimated in order to evaluate the likelihood. When estimating the dispersion, the nuisance parameters are the coefficients in the log-linear model.

This implementation calls the LAPACK library to perform the Cholesky decomposition during adjustment estimation.

The purpose of start and get.coef is to allow hot-starting for multiple calls to adjustedProfileLik, when only the dispersion is altered. Specifically, the returned GLM coefficients from one call with get.coef==TRUE can be used as the start values for the next call.

The weights argument is interpreted in terms of averages. Each value of y is assumed to be the average of n independent and identically distributed NB counts, where n is given by the weight. This assumption can generalized to fractional weights.

Value

If get.coef==FALSE, a vector of adjusted profile log-likelihood values is returned containing one element for each row of y.

Otherwise, a list is returned containing apl, the aforementioned vector of adjusted profile likelihoods, and beta, the numeric matrix of fitted GLM coefficients.

Author(s)

Yunshun Chen, Gordon Smyth, Aaron Lun

References

Cox, DR, and Reid, N (1987). Parameter orthogonality and approximate conditional inference. Journal of the Royal Statistical Society Series B 49, 1-39.

McCarthy, DJ, Chen, Y, Smyth, GK (2012). Differential expression analysis of multifactor RNA-Seq experiments with respect to biological variation. Nucleic Acids Research 40, 4288-4297. https://doi.org/10.1093/nar/gks042

See Also

glmFit

Examples

y <- matrix(rnbinom(30, mu=10, size=20), 10, 3)
design <- matrix(1, 3, 1)
dispersion <- 0.05
adjustedProfileLik(dispersion, y, design, offset=0)

Example output

Loading required package: limma
 [1]  -9.925742 -10.494508  -8.350779  -8.194301 -10.850206 -10.537687
 [7]  -7.902534  -8.922263  -8.421048 -10.023225

edgeR documentation built on Jan. 16, 2021, 2:03 a.m.