Usage: TSplit.py parameter_filename input_filename output_filename

Wrapper for the tsplit algorithm.

Note: The class labels will have the extension you specify on the command line and the tsplit intermediate file, if saved, will have a .gtr extension.

Depends on the following environment variables: TSPLIT_COMMAND (e.g., /proj/cluster_gazing2/bin/tsplit)

Brief Algorithm Description:

Required Parameters: (note: the list enclosed in the brakets are possible values each one of parameters can take )

distance_metric = [correlation, euclidean, Bhattacharyya]

Bhattacharyya : takes into account of not only the difference between the two mean vectors, but also the distributions of the two groups of data points.

agglomerate_method = [none, native, size, clusterNumber]

none - do not agglomerate, just generate the normal tsplit output files

native - use tsplit built in agglomeration to produce as close to K clusters as possible

size - perform a size threshold agglomeration. Starting at the root recurse through the tree attempting to agglomerate at each node stopping only when the number of genes in the agglomerated sub-tree is less then the parameter "size_threshold"

clusterNumber - return as close to K clusters as possible using the "size" agglomeration method to partition the tree

size/clusterNumber agglomeration is identical to the agglomeration used in xclust

splitting_method = [PCA, 'Best'}

uses either PCA splitting or best splitting, which utilizes the energy parameter

k = <x>

where x is the target number of clusters

Optional / Dependent Parameters:

min_cluster_size = <x> (required for size and native agglomeration)

where x is the minimun number of genes that will appear is any given cluster.

energy = <x> (required if method = PCA)

number in the range of (0, 100]. Indicating the quantity of energy to preserve at each node.

merge = [closest, 'prune']

Selects the method to merge node back to the target number of clusters. Only applicable if agglomerate_method is native. closest is the default and merges nodes together which are closest depending on the chosen distance_metric. prune simply merges sibling nodes togvether.

save_intermediate_files = [yes, 'no'] (default no)

if you choose yes, the generated tsplit files (.gtr) will be saved. Otherwise they will be deleted.