wallace: Computes Wallace scores comparing two clustering methods.

In tjh48/clusterSeq: Clustering of high-throughput sequencing data by identifying co-expression patterns

Description

Given two clusterings A \& B we can calculate the likelihood that two elements are in the same cluster in B given that they are in the same cluster in A, and vice versa.

Usage

wallace(v1, v2)

Arguments

v1	SimpleIntegerList object (output from makeClusters or makeClustersFF).
v2	SimpleIntegerList object (output from makeClusters or makeClustersFF).

Value

Vector of length 2 giving conditional likelihoods.

Author(s)

Thomas J. Hardcastle

Examples

# using likelihood data from a Bayesian analysis of the data

# load in analysed countData object
data(cD.ratThymus, package = "clusterSeq")

# estimate likelihoods of dissimilarity on reduced set
aM <- associatePosteriors(cD.ratThymus[1:1000,])

# make clusters from dissimilarity data
sX <- makeClusters(aM, cD.ratThymus[1:1000,], threshold = 0.5)

# or using k-means clustering on raw count data

#Load in the processed data of observed read counts at each gene for each sample. 
data(ratThymus, package = "clusterSeq")

# Library scaling factors are acquired here using the getLibsizes
# function from the baySeq package.
libsizes <- getLibsizes(data = ratThymus)

# Adjust the data to remove zeros and rescale by the library scaling
# factors. Convert to log scale.
ratThymus[ratThymus == 0] <- 1
normRT <- log2(t(t(ratThymus / libsizes)) * mean(libsizes))

# run kCluster on reduced set.
normRT <- normRT[1:1000,]
kClust <- kCluster(normRT, replicates = cD.ratThymus@replicates)

# make the clusters from these data.
mkClust <- makeClusters(kClust, normRT, threshold = 1)


# compare clusterings
wallace(sX, mkClust)

tjh48/clusterSeq documentation built on May 31, 2019, 3:40 p.m.