metaseqR (version 1.12.2)

calc.f1score: Calculate the F1-score

Description

This function calculates the F1 score (2*(precision*recall/precision+racall) or 2*TP/(2*TP+FP+FN) given a matrix of p-values (one for each statistical test used) and a vector of ground truth (DE or non-DE). This function serves as a method evaluation helper.

Usage

calc.f1score(truth, p, sig = 0.05)

Arguments

truth

the ground truth differential expression vector. It should contain only zero and non-zero elements, with zero denoting non-differentially expressed genes and non-zero, differentially expressed genes. Such a vector can be obtained for example by using the make.sim.data.sd function, which creates simulated RNA-Seq read counts based on real data. It MUST be named with gene names, the same as in p.

p

a p-value matrix whose rows correspond to each element in the truth vector. If the matrix has a colnames attribute, a legend will be added to the plot using these names, else a set of column names will be auto-generated. p can also be a list or a data frame. In any case, each row (or element) MUST be named with gene names (the same as in truth).

sig

a significance level (0 < sig

Value

• A named list with two members. The first member is a data frame with the numbers used to calculate the TP/(FP+FN) ratio and the second member is the ratio TP/(FP+FN) for each statistical test.

Examples

p1 <- 0.001*matrix(runif(300),100,3)
p2 <- matrix(runif(300),100,3)
p <- rbind(p1,p2)
rownames(p) <- paste("gene",1:200,sep="_")
colnames(p) <- paste("method",1:3,sep="_")
truth <- c(rep(1,40),rep(-1,40),rep(0,20),rep(1,10),
    rep(2,10),rep(0,80))
names(truth) <- rownames(p)
f1 <- calc.f1score(truth,p)