Evaluation metric for batch effect removal

[xuebaliang]

Hello, recently I read your paper and it is a perfect work. But I can not find the code for KL divergence of evaluation metric for batch effect removal in your paper, thus can you send me it? Thanks very much.

[eleozzr]

BatchKL=function(df,dimensionData=NULL,replicates=200,n_neighbors=100,n_cells=100,batch="BatchID"){
  #entropy of batch mixing
  #replicates is the number of boostrap times
  #n_neighbors is the number of nearest neighbours of cell(from all batchs)
  #n_cells is the number of randomly picked cells
  if (is.null(dimensionData)){
        tsnedata=as.matrix(df[,c("tSNE_1","tSNE_2")])
  }else{
        tsnedata=as.matrix(dimensionData)
  }
  batchdata=factor(as.vector(df[,batch]))
  table.batchdata=as.matrix(table(batchdata))[,1]
  tmp00=table.batchdata/sum(table.batchdata)#proportation of population
  n=dim(df)[1]
  KL=sapply(1:replicates,function(x){
    bootsamples=sample(1:n,n_cells)
    #nearest=nn2(tsnedata,tsnedata[bootsamples,],k=n_neighbors)
    nearest=nabor::knn(tsnedata,tsnedata[bootsamples,],k=min(5*length(tmp00),n_neighbors))
    KL_x=sapply(1:length(bootsamples),function(y){
      id=nearest$nn.idx[y,]
      tmp=as.matrix(table(batchdata[id]))[,1]
      tmp=tmp/sum(tmp)
      return(sum(tmp*log2(tmp/tmp00),na.rm = T))
    })
    return(mean(KL_x,na.rm = T))
  })
  return(KL)
}