clValid (version 0.6-2)

sota: Self-organizing Tree Algorithm (SOTA)

Description

Computes a Self-organizing Tree Algorithm (SOTA) clustering of a dataset returning a SOTA object.

Usage

sota(data, maxCycles, maxEpochs = 1000, distance = "euclidean", wcell = 0.01, 
     pcell = 0.005, scell = 0.001, delta = 1e-04, neighb.level = 0, 
     maxDiversity = 0.9, unrest.growth = TRUE, ...)

Arguments

data

data matrix or data frame. Cannot have a profile ID as the first column.

maxCycles

integer value representing the maximum number of iterations allowed. The resulting number of clusters returned by sota is maxCycles+1 unless unrest.growth is set to FALSE and the maxDiversity criteria is satisf

maxEpochs

integer value indicating the maximum number of training epochs allowed per cycle. By default, maxEpochs is set to 1000.

distance

character string used to represent the metric to be used for calculating dissimilarities between profiles. 'euclidean' is the default, with 'correlation' being another option.

wcell

value specifying the winning cell migration weight. The default is 0.01.

pcell

value specifying the parent cell migration weight. The default is 0.005.

scell

value specifying the sister cell migration weight. The default is 0.001.

delta

value specifying the minimum epoch error improvement. This value is used as a threshold for signaling the start of a new cycle. It is set to 1e-04 by default.

neighb.level

integer value used to indicate which cells are candidates to accept new profiles. This number specifies the number of levels up the tree the algorithm moves in the search of candidate cells for the redistribution of profiles. The default is 0.

maxDiversity

value representing a maximum variability allowed within a cluster. 0.9 is the default value.

unrest.growth

logical flag: if TRUE then the algorithm will run maxCycles iterations regardless of whether the maxDiversity criteria is satisfied or not and maxCycles+1 clusters will be produced; if FALSE then the algorithm

...

Any other arguments.

Value

• data
  data matrix used for clustering
• c.tree
  complete tree in a matrix format. Node ID, its Ancestor, and whether it's a terminal node (cell) are listed in the first three columns. Node profiles are shown in the remaining columns.
• tree
  incomplete tree in a matrix format listing only the terminal nodes (cells). Node ID, its Ancestor, and 1's for a cell indicator are listed in the first three columns. Node profiles are shown in the remaining columns.
• clust
  integer vector whose length is equal to the number of profiles in a data matrix indicating the cluster assingments for each profile in the original order.
• totals
  integer vector specifying the cluster sizes.
• dist
  character string indicating a distance function used in the clustering process.
• diversity
  vector specifying final cluster diverisities.

Details

The Self-Organizing Tree Algorithm (SOTA) is an unsupervised neural network with a binary tree topology. It combines the advantages of both hierarchical clustering and Self-Organizing Maps (SOM). The algorithm picks a node with the largest Diversity and splits it into two nodes, called Cells. This process can be stopped at any level, assuring a fixed number of hard clusters. This behavior is achieved with setting the unrest.growth parameter to TRUE. Growth of the tree can be stopped based on other criteria, like the allowed maximum Diversity within the cluster and so on. Further details regarding the inner workings of the algorithm can be found in the paper listed in the Reference section.

References

Herrero, J., Valencia, A, and Dopazo, J. (2005). A hierarchical unsupervised growing neural network for clustering gene expression patterns. Bioinformatics, 17, 126-136.

See Also

plot.sota, print.sota

Examples

data(mouse)
express <- mouse[,c("M1","M2","M3","NC1","NC2","NC3")]
rownames(express) <- mouse$ID

sotaCl <- sota(as.matrix(express), 4)
names(sotaCl)
sotaCl
plot(sotaCl)
plot(sotaCl, cl=2)