modelCalcINT {Envisage}R Documentation

Fit a model with main effect and first order interaction terms for each gene and calculate significance

Description

This is the main calculation for the package Envisage. An optimised model, containing terms shown to have practical importance for explaining variation, is produced using a step-wise procedure. Terms may be main effect terms or first order interactions. The significance of each term is computed by using a type II sum of squares.

Usage

  modelCalcINT(paramData, exprData, param4INT, startingModel,
    progress) 

Arguments

paramData An object of class phenoData containing the phenotypic information for the experiment.
exprData A matrix of expression values for each gene for each sample. Usually calculated using exprs on an object of class ExpressionSet.
param4INT A character vector specifying the variables for which first order interactions should be calculated.
startingModel An object of class formula containing the minimal model thta should be used in the step-wise model fitting process.
progress Object of class tclVar indicating progress through the genes in the gene list. Used only if widget=TRUE for envisage.

Details

The Envisage package contains methods allowing the use of linear models (LMs) for analysing significantly changing genes in experiments with a variety of sources of variation, be they experimentally controlled variables such as drug treatment or time, or non-controlled sources of confounding variation such as phenotypic or environmental differences. This allows all sources of variation to be considered when analysing for significant differential expression, ensuring resulting genes are biologically relevent to the experimental question.

The function modelCalcINT performs the main modelling step for the package Envisage. A stepwise procedure is used on a per-gene basis to calculate the optimum model for each gene. In comparison to modelCalcME, terms used in the model may be main effect terms or first-order interactions. The step-wise process begins by fitting the maximal model to the data, then removes terms using a leave-one-out approach based on the Akaike information criterion (AIC). Specification of a minimal model allows terms to be 'forced' into the final formula.

To increase speed, the model fitting procedure is first performed using only main effect terms, and first order interactions are only calculated if the main effect term is thus shown to be significant.

The significance of each term in the final model is estimated by using a Type II sum of squares to compare the model with and without each term. Also, if fitting the model results in aliased terms (either due to experimental design, insufficient data, or non-independence of model terms), this information is also returned.

modelCalcINT is an internal function and is not designed to be accessed by users.

Value

A list is returned with the following slots:

pValues An object of class data.frame containing the p-value from the Type II sum of squares analysis for each model term for each gene.
failedAnova A vector of genes for which the Anova Type II sum of squares failed, indicating model aliasing.
errorGenes A list of the genes that failed and the model terms for which aliasing was detected.

Author(s)

Sam Robson S.C.Robson@warwick.ac.uk

References

Robson, S. C., Hunter, E., Bird, H., Turner, H. (2008) Envisage: model-based significance analysis of microarray gene expression data, manuscript in preparation

See Also

For the main effect only version of this function, see modelCalcME. For the main Envisage method, see envisage.


[Package Envisage version 1.0-2 Index]