BCPS

The BCPS (Breast Cancer Purity Score) is a transcription-based score - made of 5 tumour-associated genes and 4 stroma-associated genes - that estimates the tumour content in bulk transcriptomic data from clinical breast cancer samples. Here we present a pipeline to compute the BCPS and use it to adjust gene expression for tumour purity.

Applications of BCPS

For all the BCPS applications developed so far, the programming language R and the packages singscore (version 1.14.0), stats (version 4.1.3), and Biobase (version 2.54.0) were used.

library(singscore)
library(stats)
library(Biobase)

The 9-gene BCPS is handled as a list of lists. Additional gene sets can be added to the list and evaluated together with the 9 genes in the BCPS.

We recommend using the ExpressionSet object to handle your dataset of expression profiles and upload it in R.

load("../expression_set.RData") 	#Upload the ExpressionSet object of your dataset

l.signature <- list(BCPS=list(BCPS_UP=c("AP1M2", "CDK5", "PAFAH1B3", "SLC25A10", "SMG5"),
                    BCPS_DOWN=c("CXCL12", "IFFO1", "MFAP4","TGFBR2")))

Evaluation of BCPS

To evaluate BCPS we used a function that manages the list of genesets and recognizes the expected direction of expression in case of high tumour purity (i.e. "BCPS_UP" and "BCPS_DOWN").

mySimpleScore<-function(rankData,mysetlist,knownDir=TRUE){
  require(singscore)
  score<-matrix(0, nrow=length(mysetlist),ncol=ncol(rankData))
  rownames(score)<-names(mysetlist)
  colnames(score)<-colnames(rankData)
  for(i in 1:length(mysetlist)){
    if(class(mysetlist[[i]])=="character"){
      sl<-mysetlist[[i]]
      sl<-sl[sl%in%rownames(rankData)]
      if(length(sl)>=3){
        scoretemp<-simpleScore(rankData,upSet = sl, knownDirection = knownDir)
        score[rownames(score)==names(mysetlist)[i],]<-scoretemp$TotalScore
      }
    } else {
      sl.up<-mysetlist[[i]][[grep("UP$",names(mysetlist[[i]]), ignore.case = TRUE)]]
      sl.up<-sl.up[sl.up%in%rownames(rankData)]
      sl.dn<-mysetlist[[i]][[grep("DN$|DOWN$",names(mysetlist[[i]]), ignore.case = TRUE)]]
      sl.dn<-sl.dn[sl.dn%in%rownames(rankData)]
      if(length(sl.up)>=3 & length(sl.dn)>=3){
        scoretemp<-simpleScore(rankData,upSet = sl.up, downSet = sl.dn)
        score[rownames(score)==names(mysetlist)[i],]<-scoretemp$TotalScore
      }
    }
  }
  score<-score[rowSums(score!=0)>0,]
  return(score)
}

Once the function has been defined, the BCPS can be computed:

rankData <- rankGenes(exprs(eset))
result <- mySimpleScore(rankData, l.signature)
result <- as.data.frame(result)
result$sample <- row.names(result)
colnames(result)[1] <- "BCPS"
result <- result[,c(2,1)]

In the "result" object you will find the BCPS (and other possible genesets) evaluated for each sample of your dataset.

Adjusting for sampling bias using the BCPS

To adjust for the sampling bias, the BCPS evaluated in the previous paragraph can be used to adjust the expression of each gene in your expression matrix.

expression_matrix <- exprs(eset)
patients <- colnames(expression_matrix)

adjusted_matrix <- apply(expression_matrix, 1, function(x) {
  l <- lm(x~result[patients,"BCPS"])
  x <- l$residuals})
adjusted_matrix <- as.matrix(t(adjusted_matrix))

"adjusted_matrix" contains a gene expression matrix in which each gene was adjusted based on its linear relationship with BCPS.

Publication

A manuscript is currently under consideration for publication, to cite the BCPS currently please refer to the bioRxiv preprint: https://biorxiv.org/cgi/content/short/2024.02.23.581701v1

Further information

Created by Dr Marco Barreca at the University of Milano-Bicocca in collaboration with Dr Matteo Dugo at the IRCCS San Raffaele Hospital, under the supervision of Dr Maurizio Callari at Fondazione Michelangelo.