rf_modeling: Random Forest Predictive Models

In vladpetyuk/vp.misc: Miscellaneous functions used by Vlad Petyuk for proteomics data analysis

Description

The objective is - given the number of features, select the most informative ones and evaluate the predictive random forest model. The feature and model selection performed independently for each round of LOOCV.

Usage

rf_modeling(
  msnset,
  features,
  response,
  pred.cls,
  K = NULL,
  sel.feat = T,
  sel.alg = c("varSelRF", "Boruta", "top"),
  ...
)

Arguments

msnset	MSnSet object
features	character vector features to select from for building prediction model. The features can be either in featureNames(msnset) or in pData(msnset).
response	factor to classify along. Must be only 2 levels.
pred.cls	class to predict
K	specifies the cross-validation type. Default NULL means LOOCV. Another typical value is 10.
sel.feat	logical defining if to select features or use the entire set?
sel.alg	character. varSelRF Boruta top - just selecting top 3 features
...	Extra arguments. Currently passed only to Boruta algorithm.
verbose	0 - silent, 1 - count round, 2 - print selected features at each round

Value

list

prob

is the probabilities (response) from LOOCV that the sample is "case"

features

list of selected features for each iteration of LOOCV

top

top features over all iterations

auc

AUC

pred

prediction perfomance obtained by ROCR::prediction

Examples

data(srm_msnset)
head(varLabels(msnset))
head(msnset$subject.type)
# reduce to two classes
msnset <- msnset[,msnset$subject.type != "control.1"]
msnset$subject.type <- as.factor(msnset$subject.type)
plotAUC(out)
# top features consistently recurring in the models during LOOCV
print(out$top)
# the AUC
print(out$auc)
# probabilities of classifying the sample right, if the feature selection
# and model training was performed on other samples
plot(sort(out$prob))
abline(h=0.5, col='red')

vladpetyuk/vp.misc documentation built on June 25, 2021, 6:35 a.m.