Mapping genesets for KS testing

library(BS831)
ngenes <- 10

Let us generate a toy set of 10 genes with random tScores, which we will use to establish a gene ranking and based on which we will perform the KS enrichment.

set.seed(123)
tScores <- round(rnorm(ngenes,0,1),4)
names(tScores) <- sprintf("gene%02d",1:ngenes)
print(tScores)

##  gene01  gene02  gene03  gene04  gene05  gene06  gene07  gene08  gene09  gene10 
## -0.5605 -0.2302  1.5587  0.0705  0.1293  1.7151  0.4609 -1.2651 -0.6869 -0.4457

Notice that the t.score vector is named (i.e., each score is labeled by the corresponding gene).

Let’s assume the geneset we want to test is the following:

gset <- c("gene03","gene05","gene06","gene07")

To establish the geneset members’ ranking, and to properly perform the KS test, we can proceed in one of two ways.

We can sort the t.score vector, find the position of the geneset members and call ksGenescore on those positions

## sorting the negative of the scores yields decreasing order sort
## alternatively: sortT <- tScores[order(tScores,decreasing=TRUE)]
sortT <- sort(-tScores)

## find the geneset members' positions
idx1 <- match(gset,names(sortT))
print(idx1) # show those positions (toward the end of the ranking)

## [1] 2 4 1 3

## perform ks test
ksGenescore(ngenes,idx1,do.plot=TRUE,main="by sort")

## 
##  Two-sample Kolmogorov-Smirnov test
## 
## data:  1:n.x and y
## D = 0.6, p-value = 0.2551
## alternative hypothesis: two-sided

We can compute the rank of the tScores, find the ranks of the geneset members, and call ksGenescore on those ranks

## computing ranks of minus the scores (to use decreasing order)
rankT <- rank(-tScores,ties.method="first")

## find the positions of geneset members in the rank list
idx2 <- match(gset,names(rankT))
print(rankT[idx2]) # show those ranks

## gene03 gene05 gene06 gene07 
##      2      4      1      3

## perform ks test on the ranks
ksGenescore(ngenes,rankT[idx2],do.plot=TRUE,main="by rank")

## 
##  Two-sample Kolmogorov-Smirnov test
## 
## data:  1:n.x and y
## D = 0.6, p-value = 0.2551
## alternative hypothesis: two-sided

Amy Li, Eric Reed, Stefano Monti