geNetClassifier (version 1.12.0)

queryGeNetClassifier: Queries the classifier trained with geNetClassifier.

Description

Queries the classifier trained by geNetClassifier in order to find out the class of new samples.

Usage

queryGeNetClassifier(classifier, eset, minProbAssignCoeff = 1, minDiffAssignCoeff = 0.8, verbose = TRUE)

Arguments

classifier

Classifier returned by geNetClassifier. (@classifier)

eset

ExpressionSet or Matrix. Gene expression matrix of the new samples.

minProbAssignCoeff

Numeric. Coefficient to modify the minimum probability requird to assign a sample to a class. Reduce to improve call rate. Increase to reduce error. 0: Removes this restriction. The sample will always be assigned to the class with the highest probability. between 0 and 1: Reduces the required probability to assign a sample to a class. >1: Increases the required probability. Warning: if minProbCoef is equal to 2*number of classes, all the samples will be left as 'NotAssigned'.

minDiffAssignCoeff

Numeric. Coefficient to modify the required difference between the two most likelly classes. Reduce to improve call rate. Increase to reduce error. 0: Removes this restriction. The probability of the second most-likely class will not be taken into account. between 1 and 1: Reduces the required difference to assign the sample. >1: Increases the required difference. Warning: if minDiffAssignCoeff is equal to the number of classes, all the samples will be left as 'NotAssigned'.

verbose

Logical. If TRUE, messages indicating the execution progress will be printed on screen.

Value

List:

• call Command used to execute the function.
• classes Classes to wich each of the samples were asigned to.
• probabilities Probabilities to the 2 classes each sample is most likelly to belong to.

Details

By default, in order to assign a sample two contitions must be met:

• if minProbAssignCoeff = 1The probability of belonging to the class should be at least double of the random probability.
• if minDiffAssignCoeff = 0.8The difference of probabilities between the most likely class and the second most likely class should be more than 80

  This means, that in a 4-class classifier, in order to assing a sample, the highest probabiity should be at least 0.5 (2x0.25), and the next most-likely-class should have a probability at least 0.2 (80 If these conditions are not met, the sample will be left as notAssigned.

  Modify the arguments values in order to modify these assignment conditions. Setting minProbAssignCoeff = 0 and minDiffAssignCoeff = 0 all samples will be assigned to the most likely class without any further restrictions.

See Also

Main package function and classifier training: geNetClassifier Query summary: querySummary External validation stats: externalValidation.stats and externalValidation.probMatrix

Examples

##########################
## Classifier training
##########################

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

# Select the train samples: 
# There should be the same number of samples from each class.
trainSamples<- c(1:10, 13:22, 25:34, 37:46, 49:58) 
# summary(leukemiasEset$LeukemiaType[trainSamples])

# Train a classifier or load a trained one:
# leukemiasClassifier <- geNetClassifier(leukemiasEset[,trainSamples], 
#    sampleLabels="LeukemiaType", plotsName="leukemiasClassifier") 
data(leukemiasClassifier) # Sample trained classifier

##########################
## Classifier Query
##########################
# Select the samples to query the classifier 
#   - Real use: samples whose class we want to known
querySamples <- "GSM330154.CEL"
#   - External validation: samples not used for training
querySamples <- c(1:60)[-trainSamples]         

#### Make a query to the classifier ("ask" about what class the new samples are):
queryResult <- queryGeNetClassifier(leukemiasClassifier, leukemiasEset[,querySamples])

# See the class it assigned to each sample:
queryResult$class[1:5]
# Or the samples which it wasn't sure about:
t(queryResult$probabilities[,queryResult$class=="NotAssigned"])

# Obtain an overview of the results
querySummary(queryResult)

#### Optional: Modify assignment conditions
# (minDiffCoef=0, minProbCoef=0: All samples will be assigned to the most likely class)
queryResult_AssignAll <- queryGeNetClassifier(leukemiasClassifier, 
    leukemiasEset[,querySamples], minDiffAssignCoeff=0, minProbAssignCoeff=0)
# No samples are left as "NotAssigned":
queryResult$probabilities[,queryResult_AssignAll$class=="NotAssigned"]

#### External validation:
# Confusion matrix:
confMatrix <- table(leukemiasEset[,querySamples]$LeukemiaType, 
    queryResult_AssignAll$class)
# New accuracy, call rate, sensitivity and specificity:
externalValidation.stats(confMatrix)
# Probability matrix for the assigned samples
externalValidation.probMatrix(queryResult, leukemiasEset[,querySamples]$LeukemiaType)