NAME

run_cluster.pl

A script to run the k-means cluster analysis.

SYNOPSIS

run_cluster.pl [--options...] <filename>

Options:
--in <filename>
--out <jobname> 
--num <integer>               (6)
--run <integer>               (200)
--method [a|m]                (m)
--dist [c|a|u|x|s|k|e|b]      (e)
--version
--help

OPTIONS

The command line flags and descriptions:

--in <filename>

Specify the input file. The file should be a simple tab-delimited text file, genes (features) as rows, experimental data (microarray or sequencing data) should be columns. The first column contains unique gene identifiers. A column header row is expected. Standard biotoolbox data text files with metadata lines should be exported to a compatible format using the treeview function in the manipulate_datasets.pl script. A .cdt file generated from this may also be used.

--out <jobname>

Specify the output jobname, which will be the basename of the output files. By default it uses the input base filename.

--num <integer>

Specify the number of clusters to identify. Default value is 6.

--run <integer>

Enter the number of iterations to run the cluster algorithm to find an optimal solution. The default value is 500.

--method [a|m]

Specify the method of finding the center of a cluster. Two values are allowed, arithmetic mean (a) and median (m). Default is mean.

--dist [c|a|u|x|s|k|e|b]

Specify the distance function to be used. Several options are available.

c  correlation
a  absolute value of the correlation
u  uncentered correlation
x  absolute uncentered correlation
s  Spearman's rank correlation
k  Kendall's tau
e  Euclidean distance
b  City-block distance

The default value is 'e', Euclidean distance.

--version

Print the version number.

--help

Display this POD documentation.

DESCRIPTION

This program is a wrapper around the Cluster 3.0 C library, which identifies clusters between genes. Currently the program performs the k-means or k-medians functions, although other functions could be implemented if requested.

Please refer to the Cluster 3 documentation for more detailed information regarding the implementation and detailed methods. Documentation may be found at http://bonsai.hgc.jp/~mdehoon/software/cluster/.

Select the desired number of clusters that are appropriate for your dataset and an appropriate number of iterations. The default values are fine to start with, but should be customized for your dataset. In general, empirically test a range of cluster numbers, e.g. 2 to 12, to find the optimal cluster number that is both informative and manageable. Increasing the number of iterations will increase confidence at the expense of compute time. The goal is to find an optimal solution more than once; the more times a solution has been found, the higher the confidence. Note that noisy or very large datasets may never yield more than 1 solution.

The resulting CDT files may be visualized using the Java Treeview program, found at http://jtreeview.sourceforge.net.

AUTHOR

Timothy J. Parnell, PhD
Howard Hughes Medical Institute
Dept of Oncological Sciences
Huntsman Cancer Institute
University of Utah
Salt Lake City, UT, 84112

This package is free software; you can redistribute it and/or modify it under the terms of the Artistic License 2.0.