NetWeaver (version 0.0.6)

ensemble_rank: Ensemble Rank

Description

Compute composite rank score using evidence from multiple features.

Usage

ensemble_rank(x, method=c('ProductOfRank','MeanOfLog','MeanOfLogLog','Mean'),
  small=1.0e-320, standardize=TRUE)

Arguments

x

a matrix of discriminant values (e.g., P values) measuring the strength of the association between objects (eg modules) (in rows) and variables/features (in columns).

method

a character string specifyign the ranking metric. See Details.

small

offset the small p values before taking log transformation.

standardize

whether to rescale the final ranking score by dividing the maximum value.

Value

A vector of ranking scores. The larger the score, the higher the rank.

Details

Three ranking metrics are currently implemented:

ProductOfRank is the one used in Zhang et al. (Cell 2013, 153: 707-720) with a slight change to standardize the scores to be between 0 and 1;

MeanOfLog computes the mean of -log(p value) which penalizes insignificant p values;

MeanOfLogLog computes the mean of log(-log(p value)) which penalizes insignificant p values and shrinks the difference in the ranges of p values.

Means computes the row mean of absolute values of x.

Examples

# NOT RUN {
#Rank US cities by the significance test of excessive arrests compared to average
#compute the p value of one-tailed z-test
x=apply(USArrests,2,function(x) pnorm((x-mean(x))/sd(x),lower.tail=FALSE))
#compute ranking score using three different metrics
score1=ensemble_rank(x,method='ProductOfRank')
score2=ensemble_rank(x,method='MeanOfLog')
score3=ensemble_rank(x,method='MeanOfLogLog')
#plot the top 5 worst cities by each metric
par(mfrow=c(1,3))
barplot(sort(score1,decreasing=TRUE)[1:5],las=2)
barplot(sort(score2,decreasing=TRUE)[1:5],las=2)
barplot(sort(score3,decreasing=TRUE)[1:5],las=2)
# }