README.md

In jweile/yogiroc: Draw ROC and PRC curves

yogiroc

YogiROC is an R package that allows for convenient plotting of Receiver-Operator Characteristics (ROC) and Precision-Recall Curves (PRC), including plotting of confidence intervals, significance testing between area under the PRC curve (AUPRC) and prior-balancing.

Author contributions:

• Prior-balancing of PRC curves was conceptualized by Yingzhou Wu and Frederick P Roth (Wu et al. unpublished).
• Significance testing of AUPRC differences and confidence intervals were conceptualized by Jochen Weile.
• R implementation by Jochen Weile.

Installation

Installation requires R version 3.6 or higher.

remotes::install_github("jweile/yogiroc")

Examples:

#install library
devtools::install_github("jweile/yogiroc")

#load library
library(yogiroc)

#create some fake example data
fakeData <- data.frame(
  pathogenic = c(rep(TRUE,90),rep(FALSE,60)),
  prediction1 = c(rnorm(90,mean=2,sd=1),rnorm(60,mean=1,sd=1)),
  prediction2 = c(rnorm(90,mean=1.5,sd=1),rnorm(60,mean=1,sd=1))
)

#create yogiroc2 object
yrobj <- yr2(truth=fakeData$pathogenic, scores=fakeData[,-1])

#draw PRC curves
draw.prc(yrobj)

#draw prior-balanced PRC curves
draw.prc(yrobj,balanced=TRUE)

#use custom colors and other graphical parameters
draw.prc(yrobj,col=c("chartreuse3","firebrick3"),main="My PRC curve")

#draw PRC curves without monotonization
draw.prc(yrobj,monotonized=FALSE)

#draw PRC curves with confidence itervals
draw.prc.CI(yrobj)

#calculate AUPRC / AUBPRC / R90P
auprc(yrobj)
auprc(yrobj,balanced=TRUE)
recall.at.prec(yrobj,0.9,balanced=TRUE)

#calculate whether differences in AUPRC are significant
aucSignif <- auprc.signif(yrobj)
sprintf("Prediction 1 has an AUPRC of %.02f with confidence interval [%.02f;%.02f].",
  aucSignif$auprc[[1]], aucSignif$ci[1,1], aucSignif$ci[2,1]
)
sprintf("Prediction 2 has an AUPRC of %.02f with confidence interval [%.02f;%.02f].",
  aucSignif$auprc[[2]], aucSignif$ci[1,2], aucSignif$ci[2,2]
)
sprintf("Prediction 1 is better than Prediction 2, with p-value=%.03f and LLR=%.02f",
   aucSignif$pval[1,2], aucSignif$llr[1,2]
)

jweile/yogiroc documentation built on Jan. 15, 2024, 2:47 a.m.