BRS: Main BRS subclass predictor

In CostelloLab/BRSpred: BCG Response Subtype Predictor for High-Risk Non-Muscle Invasive Bladder Cancer

Main BRS subclass predictor

Description

Main BRS subclass predictor

Usage

BRS(
  newx,
  trainx = as.matrix(BRSpred::CohortA_pre),
  trainy = BRSpred::erasmus_clinical[colnames(BRSpred::CohortA_pre), "Erasmus.BRS"],
  common = TRUE,
  genes,
  pamrobj,
  seed = 1234,
  nfold,
  threshold,
  type = "class",
  scale = c("together", "independent", "none"),
  qnormalize = TRUE,
  impute = FALSE,
  getall = FALSE,
  verb = TRUE,
  ...
)

Arguments

newx	Input data matrix
trainx	Training data matrix x; by default Cohort A pre-BCG samples
trainy	Training subclasses y; by default BRS classes identified via consensus clustering for Cohort A pre-BCG samples
common	Should only common genes be considered (prevents need for potential imputation)
genes	Subset of genes used in the training; by default top 2000 variable genes in the training data
pamrobj	Previously trained pamr-object used for prediction; by default pamr object is instead trained on the fly based on trainx and trainy
seed	RNG seed, if cross-validation is used for threshold-determination this should be set; by default 1234
nfold	If provided, will designate number of folds in the CV (defaults to pamr.cv's default value)
threshold	pamr-prediction threshold parameter (if missing, by default the optimal threshold is identified by minimizing CV misclassification rate)
type	Type of prediction as given by pamr; eligible values 'class', 'posterior', 'centroid', 'nonzero' (see ?pamr.predict)
scale	Z score based scaling of data; none, or together with or independently of the training data; notice that the centroids are sensitive to scaling
qnormalize	Should quantile normalization be applied to 'x' in respect to the training data (by default TRUE)
impute	Should gene imputation be allowed via internal function if not pre-processed by user (by default FALSE)
getall	Should all objects be returned - will instead create a list with predictions, pamr object, newx, trainx, and trainy (by default FALSE)
verb	Verbosity
...	Additional parameters passed on to pamr::pamr.predict

Details

This is the main BCG response subtype predictor function. It essentially works as a wrapper for training a 3-class pamr-object classifier, with optimal threshold determined using cross-validation. Convenience functions, such as z-score scaling, quantile normalization, and missing gene imputation are provided. Notice however, that the centroid based classifier may be sensitive to the choices in respect to these parameters.

Value

pamr::pamr.predict-call predictions (getall == FALSE) or a list with the prediction and corresponding data matrices and pamr-object and threshold (getall == TRUE)

References

de Jong F. C., Laajala T. D., et al. Citation

Examples

library(BRSpred)
predict_post <- BRSpred::BRS(newx = BRSpred::CohortA_post, scale = "together")
predict_cohortb <- BRSpred::BRS(newx = BRSpred::CohortB, scale = "independent")

CostelloLab/BRSpred documentation built on Aug. 14, 2024, 9:17 a.m.