In bhklab/mCI: Modified Concordance Index

knitr::opts_chunk$set(echo = TRUE)
require(devtools)

Introduction

The concordance index (CI) is a widely used to assess the predictive value of biomedical assays. CI has previously been used in cell line-based drug screening to estimate the probability that two randomly-chosen cell lines are ranke identically across biological replicates or independent studies based on their drug response. However, the conventional formulation of CI does not account for the high level noise in drug response measurements. Here we introduce the wCI package implementing a modification of the concordance index (CI), called the robust concordance index (rCI) to explicitly accounts for the intrinsic variability of the pharmacological assays.

Installing the package

devtools::install_github("bhklab/wCI")

Using the package

Load library

library(wCI)

Load real data example. In this example we load the responses of applying PLX4720 drug in two big pharmacogenomic datasets (CTRPv2 and GDSC) on the same cell lines.

load("../data/PLX4720_data.rda")

Below is a plot of the Area-Above-the-dose-response-Curve of all points in the experiment mentined above.

plot(PLX4720_data[,"AAC_CTRPv2"],PLX4720_data[,"AAC_GDSC"], 
  main = "AAC drug response consistency between CTRPv2 and GDSC\nPLX4720 drug 
  as an example", xlab = "CTRPv2[AAC]", ylab = "GDSC[AAC]", xlim = c(0,0.5),
  ylim = c(0,0.5))

If we calculate the regular concordance index, we find that the two datasets disagree a lot!

paired.concordance.index(predictions = PLX4720_data[,"AAC_CTRPv2"],
                         observations = PLX4720_data[,"AAC_GDSC"],
                         delta.pred = 0, 
                         delta.obs = 0,
                         outx = TRUE)$cindex

However, if we account for the noise when replicating the experiments and use wCI, we are able to capture the true similarity between the experiments in the two datasets

paired.concordance.index(predictions = PLX4720_data[,"AAC_CTRPv2"], 
                          observations = PLX4720_data[,"AAC_GDSC"],
                          delta.pred = 0.2, 
                          delta.obs = 0.2,
                          outx = TRUE)$cindex