BCPlaid {biclust}

The Plaid Model Bicluster algorithm

Description

Performs Plaid Model Biclustering as described in Turner et al., 2003. This is an improvement of original 'Plaid Models for Gene Expression Data' (Lazzeroni and Owen, 2002). This algorithm models data matrices to a sum of layers, the model is fitted to data through minimization of error.

Usage

## S4 method for signature 'matrix,BCPlaid'
biclust(x, method=BCPlaid(), cluster="b", fit.model = y ~ m + a + b,
  background = TRUE, background.layer = NA, background.df = 1, row.release = 0.7, 
  col.release = 0.7, shuffle = 3, back.fit = 0, max.layers = 20, iter.startup = 5,
  iter.layer = 10, verbose = TRUE)

Arguments

x

The data matrix where biclusters have to be found

method

Here BCPlaid, to perform Plaid algorithm

cluster

'r', 'c' or 'b', to cluster rows, columns or both (default 'b')

fit.model

Model (formula) to fit each layer. Usually, a linear model is used, that estimates three parameters: m (constant for all elements in the bicluster), a(contant for all rows in the bicluster) and b (constant for all columns). Thus, default is: y ~ m + a + b.

background

If 'TRUE' the method will consider that a background layer (constant for all rows and columns) is present in the data matrix.

background.layer

If background='TRUE' a own background layer (Matrix with dimension of x) can be specified.

background.df

Degrees of Freedom of backround layer if background.layer is specified.

shuffle

Before a layer is added, it's statistical significance is compared against a number of layers obtained by random defined by this parameter. Default is 3, higher numbers could affect time performance.

iter.startup

Number of iterations to find starting values

iter.layer

Number of iterations to find each layer

back.fit

After a layer is added, additional iterations can be done to refine the fitting of the layer (default set to 0)

row.release

Scalar in [0,1](with interval recommended [0.5-0.7]) used as threshold to prune rows in the layers depending on row homogeneity

col.release

As above, with columns

max.layers

Maximum number of layer to include in the model

verbose

If 'TRUE' prints extra information on progress.

Value

Returns an Biclust object.

Author(s)

Adaptation of original code from Heather Turner from Rodrigo Santamaria [e-mail]. [e-mail]

References

Heather Turner et al, "Improved biclustering of microarray data demonstrated through systematic performance tests",Computational Statistics and Data Analysis, 2003, vol. 48, pages 235-254.

Lazzeroni and Owen, "Plaid Models for Gene Expression Data", Standford University, 2002.

Examples

  #Random matrix with embedded bicluster
  test <- matrix(rnorm(5000),100,50)
  test[11:20,11:20] <- rnorm(100,3,0.3)
  res<-biclust(test, method=BCPlaid())
  res

  #microarray matrix
  data(BicatYeast)
  res<-biclust(BicatYeast, method=BCPlaid(), verbose=FALSE)
  res