Class KMeans_Clustering

Object

Clustering

KMeans_Clustering

public class KMeans_Clustering
extends Clustering

An instance of the KMeans_Clustering class represents a clustering (i.e., grouping or partitioning) of a collection of genes using the k-means clustering method.

Field Summary

Fields inherited from class Clustering

clusters, experiments, genes

Constructor Summary

Constructors

Constructor and Description
KMeans_Clustering(String fileName, int numClusters) Creates an initially empty KMeans_Clustering.

Method Summary

Modifier and Type	Method and Description
boolean	assignGenesToClusters(java.util.Vector<java.util.Vector<Double>> means) Assigns each gene to the cluster whose mean expression vector is closest to the gene.
java.util.Vector<java.util.Vector<Double>>	getMeansOfAllClusters() Return a collection of the mean (average) expression vectors for all of the clusters.
void	initializeAllClusters() Initializes each cluster so that each cluster contains zero genes.
void	kMeans() Performs k-means clustering.
static void	main(String[] args) The main method creates a Clustering based on gene and experiment data from a tab-delimited text file.
void	populateEmptyClusters() If any clusters are empty (contain zero genes), then genes are moved from clusters containing multiple genes.
void	randomlyAssignGenesToClusters() Assigns each gene to a random cluster.

Methods inherited from class Clustering

getExperimentNamesFromFile, getGeneInformationFromFile, getNumClusters, getNumExperiments, getNumGenes, toString

Methods inherited from class Object

clone, equals, finalize, getClass, hashCode, notify, notifyAll, wait, wait, wait

Constructor Detail

KMeans_Clustering

public KMeans_Clustering(String fileName,
                         int numClusters)

Creates an initially empty KMeans_Clustering.

A set of genes and experiments are determined from the specified String representing the name of a file. Genes and experiments are read-in from the tab-delimited file. Initially, the constructed KMeans_Clustering consists of the specified number of clusters, but each cluster has not yet been assigned any genes.

Parameters:

fileName - the name of a tab-delimited text file containing gene and experiment data

numClusters - an integer representing the desired number of clusters

Method Detail

initializeAllClusters

public void initializeAllClusters()

Initializes each cluster so that each cluster contains zero genes.

getMeansOfAllClusters

public java.util.Vector<java.util.Vector<Double>> getMeansOfAllClusters()

Return a collection of the mean (average) expression vectors for all of the clusters.

Returns:

a collection of Vectors such that each Vector in the collection represents a collection of expression values

populateEmptyClusters

public void populateEmptyClusters()

If any clusters are empty (contain zero genes), then genes are moved from clusters containing multiple genes.

Assumes all genes have been assigned to a cluster. For any empty clusters, a cluster with mulitple genes is found randomly and one gene is removed from the multiple gene cluster and added to the empty cluster. When the method completes, no cluster contains zero genes.

randomlyAssignGenesToClusters

public void randomlyAssignGenesToClusters()

Assigns each gene to a random cluster.

Each gene is randomly assigned to one of the clusters. When the method completes, no cluster should contain zero genes.

assignGenesToClusters

public boolean assignGenesToClusters(java.util.Vector<java.util.Vector<Double>> means)

Assigns each gene to the cluster whose mean expression vector is closest to the gene.

The parameter means corresponds to a collection of Vectors (analagous to a 2D array). Each entry of means corresponds to the set of average expression values for a particular cluster. The method returns true if the cluster assignments are not improving, i.e., we have achieved a locally optimal clustering. The method returns false if the cluster assignments are an improvement over previous cluster assignments, i.e., better clusters than previously have been found. The measure used for comparing one clustering to another clustering is the sum of the distance of each gene from its cluster's mean.

Parameters:

means - a collection of Vectors where each Vector corresponds to a set of expression values

Returns:

a boolean indicating if the cluster assignments have improved (false) or not (true)

kMeans

public void kMeans()

Performs k-means clustering.

Initially, genes are randomly assigned to clusters. Then, iteratively, the clustering is improved until a local optima is reached (until clusterings are no longer improving). In each iteration, first the mean (average) expression vector is calculated for each cluster. Second, each gene is assigned to the cluster whose mean (average) expression vector is closest to the gene's expression vector.

main

public static void main(String[] args)

The main method creates a Clustering based on gene and experiment data from a tab-delimited text file.

The clustering is determined using k-means clustering. The text file and the desired number of clusters are specified as command line arguments. The computed set of clusters is output to standard output (and can be redirected to a file).

Parameters:

args - an array of Strings representing any command line arguments