Learn R Programming

geNetClassifier (version 1.12.0)

geNetClassifier: Main function of the geNetClassifier package. Trains the multi-class SVM classifier based on the given gene expression data through transparent detection of gene markers and their associated networks.

Description

Allows to train the classifier, calculate the genes network...

Usage

geNetClassifier(eset, sampleLabels, plotsName = NULL, buildClassifier = TRUE, estimateGError = FALSE, calculateNetwork = TRUE, labelsOrder = NULL, geneLabels = NULL, numGenesNetworkPlot = 100, minGenesTrain = 1, maxGenesTrain = 100, continueZeroError = FALSE, numIters = 6, lpThreshold = 0.95, numDecimals = 3, removeCorrelations = FALSE, correlationsThreshold = 0.8, correlationMethod = "pearson", removeInteractions = FALSE, interactionsThreshold = 0.5, minProbAssignCoeff = 1, minDiffAssignCoeff = 0.8, IQRfilterPercentage = 0, skipInteractions = TRUE, precalcGenesNetwork = NULL, precalcGenesRanking = NULL, returnAllGenesRanking = TRUE, verbose = TRUE)

Arguments

eset
ExpressionSet or matrix. Gene expression of the train samples (positive & non-logaritmic normalized values).
sampleLabels
Character. PhenoData variable (column name) containing the train samples class labels. Matrix or Factor. Class labels of the train samples.
labelsOrder
Vector or Factor. Order in which the labels should be shown in the returned results and plots.
plotsName
Character. File name with which the plots should be saved. If not provided, no plots will be drawn.
buildClassifier
Logical. If TRUE trains a classifier with the given samples.
estimateGError
Logical. If TRUE uses cross-validation to estimate the Generalization Error of a classiffier trained with the given samples.
calculateNetwork
Logical. If TRUE calculates the coexpression network between the best genes.
geneLabels
Vector or Matrix. Gene name, ID or label which should be shown in the returned results and plots.
numGenesNetworkPlot
Integer. Number of genes to show in the coexpression network for each class.
minGenesTrain
Integer. Maximum number of genes per class to train the classifier with.
maxGenesTrain
Integer. Maximum number of genes per class to train the classifier with.
continueZeroError
Logical. If TRUE, the program will continue testing combinations with more genes even if error 0 has been reached.
numIters
Integer. Number of iterations to determine the optimum number of genes (between minGenesTrain and maxGenesTrain).
lpThreshold
Numeric between 0 and 1. Required posterior probability value to consider a gene 'significant'.
removeCorrelations
Logical. If TRUE, no correlated genes will be chosen to train the classifier.
correlationsThreshold
Numeric between 0 and 1. Minimum Pearson's correlation coefficient to consider genes correlated.
correlationMethod
"pearson", "kendall" or "spearman". Type of correlation to calculate between genes.
removeInteractions
Logical. If TRUE, genes with Mutual Information coefficient over the threshold will not be chosen to train the classifier.
interactionsThreshold
Numeric between 0 and 1. Minimum Mutual Information coefficient to consider two genes equivalent.
numDecimals
Integer. Number of decimals to show in the statistics.
minProbAssignCoeff
Numeric. Allows modifying the required probability to assign a sample to a class in the internal crossvalidation. For details see: queryGeNetClassifier
minDiffAssignCoeff
Numeric. Allows modifying the difference of probabilities required between the most likely class and second most likely class to assign a sample. For details see: queryGeNetClassifier
IQRfilterPercentage
Integer. InterQuartile Range (IQR) filter applied to the initial data. Not recommended for more than two classes.
skipInteractions
Logical. If TRUE, the interactions between genes are not calculated (they will not appear on the genes network). Saves some execution time. Only available if removeInteractions=FALSE.
precalcGenesNetwork
GenesNetwork from a previous execution with the same expression data and parameters.
precalcGenesRanking
GenesRanking from a previous execution with the same expression data and parameters.
returnAllGenesRanking
Logical. If TRUE, returns the whole genes ranking. If FALSE the returned ranking contains only the significant genes (genes over lpThreshold).
verbose
Logical. If TRUE, messages indicating the execution progress will be shown.

Value

A GeNetClassifierReturn object containing the classifier and the genes chosen to train it (classificationGenes), Cross-Validation statistics, the whole GenesRanking and each class' GenesNetwork (if requested). Several plots saved as 'plotsName_....pdf' in the working directory.

References

Packages used by this function: EBarrays: emfit (Implements EM algorithm for gene expression mixture model) and ebPatterns, for calculating the gene ranking. Ming Yuan, Michael Newton, Deepayan Sarkar and Christina Kendziorski (2007). EBarrays: Unified Approach for Simultaneous Gene Clustering and Differential Expression Identification. R package. e1071: svm. Evgenia Dimitriadou, Kurt Hornik, Friedrich Leisch, David Meyer and Andreas Weingessel (2011). e1071: Misc Functions of the Department of Statistics (e1071), TU Wien. R package. http://CRAN.R-project.org/package=e1071

ipred: kfoldcv (computes feasible sample sizes for the k groups in k-fold cv) for the cross-validations. Andrea Peters and Torsten Hothorn (2012). ipred: Improved Predictors. R package. http://CRAN.R-project.org/package=ipred

minet for the Mutual Information network. Patrick E. Meyer, Frederic Lafitte and Gianluca Bontempi (2008). MINET: An open source R/Bioconductor Package for Mutual Information based Network Inference. BMC Bioinformatics. http://www.biomedcentral.com/1471-2105/9/461

RColorBrewer for palettes in some of the plots. Erich Neuwirth (2011). RColorBrewer: ColorBrewer palettes. R package. http://CRAN.R-project.org/package=RColorBrewer

igraph for the graphical representation of the networks. Csardi G, Nepusz T: The igraph software package for complex network research, InterJournal, Complex Systems 1695. 2006. http://igraph.sf.net

See Also

To query the classifier: queryGeNetClassifier All functions in the package: geNetClassifier-package

Examples

Run this code
########
# Load libraries and training data
########

# Load an expressionSet:
library(leukemiasEset)
data(leukemiasEset)

# Select the train samples: 
trainSamples<- c(1:10, 13:22, 25:34, 37:46, 49:58) 
# summary(leukemiasEset$LeukemiaType[trainSamples])


########
# Training
########

# NOTE: Training the classifier takes a while... 
# Choose ONE of the followings, or modify to suit your needs:
## Not run: 
# 	
# # "Basic" execution: All default parameters
# leukemiasClassifier <- geNetClassifier(eset=leukemiasEset[,trainSamples], 
#     sampleLabels="LeukemiaType", plotsName="leukemiasClassifier") 
# 
# # All default parameters also estimatings the classiffier's Generalization Error:
# # ( by default:  buildClassifier=TRUE, calculateNetwork=TRUE)
# # Takes longer time than the basic execution
# leukemiasClassifier <- geNetClassifier(eset=leukemiasEset[,trainSamples], 
#     sampleLabels="LeukemiaType", plotsName="leukemiasClassifier",
#     estimateGError=TRUE) 
# 
# # Faster execution (few minutes - depending on the computer): 
# # By skipping the calculation of the interactions (MI) betwen the genes, 
# # and reducing the number of genes to explore when training the classifier 
# # (100 by default), the execution time can be sightly reduced
# leukemiasClassifier <- geNetClassifier(eset=leukemiasEset[,trainSamples],
# sampleLabels="LeukemiaType", plotsName="leukemiasClassifier", 
# skipInteractions= TRUE, maxGenesTrain=20)
# 
# # To any of these examples, you can add/remove the argument geneLabels,
# # in order to show/remove the gene name in the rankings and plots:
# # The argument labelsOrder allows showing the classes in a specific order
# # i.e.: labelsOrder=c("ALL","CLL","AML",CML","NoL")
# 
# save(leukemiasClassifier, file="leukemiasClassifier.RData")  # Save execution result 
# # For loading the saved object in the future... 
# # (If it doesn't find it, use getwd() to make sure you are in the right directory)
# #load("leukemiasClassifier.RData")										
# 
# 
# # To avoid having to train a classifier to continue learning to use the package, 
# # you can load the package's pre-executed example:
# data(leukemiasClassifier)  					
# #This example classifier was trained with the following code:
# #leukemiasClassifier <- geNetClassifier(leukemiasEset[,trainSamples], 
# #    "LeukemiaType", plotsName="leukemiasClassifier", buildClassifier=TRUE, 
# #    estimateGError=TRUE, calculateNetwork=TRUE, geneLabels=geneSymbols)
# 
# ########
# # Explore the returned object:
# ########
# names(leukemiasClassifier)
# # More details on the class' help file:
# ?GeNetClassifierReturn
# 
# # Further options:
# # The trained classifier can be used to find the class of new samples:
# ?queryGeNetClassifier
# 
# # The default plots can be modified and presonalized to fit the user needs: 
# ?calculateGenesRanking
# ?plotNetwork
# ?plotDiscriminantPower
# ?plotExpressionProfiles
# ## End(Not run)

Run the code above in your browser using DataLab