getFinalDispersions: Estimate Negative Binomial dispersion paramteter taking into...

In eachanjohnson/concensusGLM: concensusGLM

Description

Estimate Negative Binomial dispersion paramteter taking into account experimental batch effects.

Usage

getFinalDispersions(x, ...)

## Default S3 method:
getFinalDispersions(x, ...)

## S3 method for class 'concensusWorkflow'
getFinalDispersions(x, ...)

## S3 method for class 'concensusDataSet'
getFinalDispersions(x, max_rows = 10000, ...)

Arguments

x	concensusWorkflow or concensusDataSet.
...	Other arguments.
max_rows	Numeric. Maximum number of observations to use for MLE.

Details

Uses the CR penalized maximum profile likelihood method, holding the μ of a GLM fixed and finding the optimal dispersion α using a Newton-type algorithm as implemented in nlm.

If the predicted_null_count column is present in the data attribute of concensusDataSet, it is added to the GLM as an offset. If the batch effects are real, this should raise the likelihood of and shrink the size of the final dispersion parameter.

This method will find a dispersion value with and without taking into account predicted_null_count, saving both results to the columns of the dispersion attribute of concensusDataSet.

Value

concensusWorkflow or concensusDataSet with a new small_model_dispersion and a new full_model_dispersion column in the dispersion attribute.

See Also

nlm, glm

eachanjohnson/concensusGLM documentation built on June 26, 2019, 2:26 a.m.