snv.performance: snv.performance

In mrrichowen/snv.predict: snv.predict uses true and false positives to improve snvs called from single cell RNA-seq data.

Description

This function takes the output files from snv.benchmark, makes a snv prediction model and tests it on hold out data. The output is a confusion matrix, a set of snv predictions and a table for plotting a ROC curve (all applied to test data only, AUC is output in STDOUT).

Usage

snv.performance(snv_data, classifiers = NULL, test_index = NULL,
  threshold = NULL, model.method = "randomforest")

Arguments

snv_data	Output from snv.benchmark. A data frame containing all derived snv classifiers with ground truth data appended.
classifiers	Defaults to use all classifiers. We recommend providing a vector of the low correlating classifiers to use in building the prediction model. This is already output from snv.bench.
test_index	A vector to determine which samples to train the model on (1) and then test (2). This can be user input to allow consistency between modelling methods if required.
threshold	Defaults to 0.5. Values 0 to 1 can be input to tune predictions for specificity or sensitivity.
model.method	Defaults to standard random forest, other options are "tuned_randomforest", "svm", and "tuned_svm".

Examples

snv.performance(snv_data,classifiers,test_index=NULL,threshold=NULL,model.method="randomforest")

mrrichowen/snv.predict documentation built on May 14, 2019, 5:27 p.m.