| Getting Started with MatArray | Search Reference |
Clustering
Clustering, especially hierarchical clustering, has proven itself one of the most useful tool to deal with microarray data. The toolbox offers efficient implementations of the most usual algorithms.Hierachical clustering
The most common framework to analyze microarray data is hierarchical clustering. The toolbox offers a fast implementation of these algorithms as a mex-file,hierarc.
The position of the "left" and "right" arms at each node
in a hierarchical clustering is arbitrary. One way to choose
a clustering is to search for the ordering for which
the sum of the distances between adjacent leaves is minimal.
This can be calculated using the mex-file
orderleaves.
If you have the statistic toolbox installed and want to
use clustering generated in one toolbox on the other, two
functions are available for the translation:
clustM2S and
clustS2M.
Visualization
A hierarchical clustering can be visualized in two different ways: in the command window, for a quick-and-dirty first look usingdisptree,
or it can be exported for visualization with TreeView using
clustTV.
TreeView is the program developed by Michael Eisen which was used to generate the typical microarray pictures we all have seen. It can be found here.
K-means clustering
A classical clustering algorithm is the K-means clustering. An efficient implementation of this algorithm has been done, with a small quirk added for better results. See the reference onkmeans for details.