MLGWAS-class: Machine Learning for Genome Wide Association Studies (MLGWAS)

In olivmrtn/MachineLearningGWAS: Machine Learning for Genome Wide Association Studies

Description

The MLGWAS contains one or more classifiers with information about their global performance, local errors and feature importance. It is meant to simplify machine learning in genome-wide association studies.

Arguments

x	MLGWAS object.

Slots

performances

Global performance measures of classifiers. First two columns are ClassifierIDs and PartitionIDs. Other columns are performance measures. For categorical responses, these are: sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), ROC plot AUC and F1 score. For continuous respones, these are: root-mean-square deviation (RMSD) and coefficient of determination (r2).

featimp

Variable importance of FeatureID as estimated by varImp. Columns are ClassifierID, PartitionID, FeatureID and Importance.

errors

Errors per partition and per sample. Columns are ClassifierID, PartitionID, SampleID and Error. For categorical variables, values are integers: 1 correspond to a mistake, 0 to a success. For continuous, values the square of the difference between the prediction and the actual response. In both cases, NA means that this sample was not part of the test set.

classifiers

Classifiers. Coded as a list of lists of train objects. First level corresponds to ClassifierID and second level to PartitionID.

partitions

Train and test set partitions Coded as a list of lists. First level corresponds to ClassifierIDs and second level to PartitionIDs.

predictions

Predictions of response(mdt) for test set. Coded as a list of lists. First level corresponds to ClassifierIDs and second level to PartitionIDs.

call

List of calls. Names of list are ClassifierIDs

Constructors

trainClassifier trains a classifier, using an MDT object, that that predicts response(mdt) affection using mtable(mdt) or phenotype(mdt) or both.

trainClassifiers allows to call trainClassifier using different combination of parameters specified as a data.frame.

Identifiers

Four identifiers are present in this class:

• ClassifierID as specified by the id argument in trainClassifier.

• PartitionID as specified by the names of the output of the partition argument in trainClassifier.

• FeatureID as specified by the FeatureIDs of the MDT object.

• SampleID as specified by the Samples of the MDT object.

Subset and Combine

x[ClassifierIDs] subsets MLGWAS objects. Must be a character.

bindML(...) combines MLGWAS objects.

Summaries

The following functions summarize results throughout partitions:

summaryPerf summarizes performance per classifier.

summaryFeatimp summarizes feature importance per classifier.

summaryErrors summarizes errors per classifier and per sample.

Plots

plotPerf plots performance as boxplots.

plotFeatimp plots feature importance as boxplots.

plotErrors plots sample error as boxplots.

manhattanPlot creates a Manhattan plot using the variable importance of features as estimated by varImp instead of -log10 pvalues.

Accessors

performances(x) and performances(x) <- value gets or sets performances.

featimp(x) and featimp(x) <- value gets or sets featimp.

errors(x) and errors(x) <- value gets or sets errors.

classifiers(x) and classifiers(x) <- value gets or sets classifiers.

partitions(x) and partitions(x) <- value gets or sets partitions.

predictions(x) and predictions(x) <- value gets or sets predictions.

classifierNames(x) and classifierNames(x) <- value gets or sets ClassifiersIDs.

partitionNames(x) and partitionNames(x) <- value get or sets PartitionIDs.

features(x) and features(x) <- value get or sets FeatureID.

samples(x) and samples(x) <- value get or sets SampleIDs

See Also

trainClassifier trainClassifiers MDT train data.table

olivmrtn/MachineLearningGWAS documentation built on May 24, 2019, 12:52 p.m.