parAttractorScanning: Parallelized attractor scanning function
In weiyi-bitw/cafr: Attractor Metagenes Finding Algorithm

Description Usage Arguments Details Value Note Author(s) References See Also Examples

Run attractor scanning on a list of seed genes.

1	parAttractorScanning(data, taskList=list(1:nrow(data)), wid=1, a=5, maxIter=100, epsilon=1E-14, bin=6, so=3, rankBased=FALSE, negateMI=TRUE)

`data`	An expression matrix with genes in the rows, samples in the columns.
`taskList`	List of vectors contains indices of seed genes to be run by each worker.
`wid`	Worker ID to be assinged the tasks.
`a`	Exponent of the mutual information, used to create weight vector for metagenes.
`maxIter`	Max number of iterations.
`epsilon`	Threshold of convergence.
`bin`	Number of bins used when estimate mutual information (default=6).
`so`	Spline order used when estimate mutual information (default=3).
`rankBased`	When `TRUE`, convert the expression values into ranks.
`negateMI`	When `TRUE`, negate the mutual information if the two vectors have negative momentum.

parAttractorScanning performs CAFrun on a subset of seeds in the dataset. The function takes the tasks as assigned to its worker ID. When applied on a parallel framework such as Sun Grid Engine, it becomes a parallelized version of attractorScanning. The function also reduces the size of seed list in the same way as attractorScanning does, so it is recommended that each worker should be given at least 100 seed genes for efficiency.

The scripts for running parallelized attractor scanning are given in "cafr/parallel" see "README.txt" for more information.

Returns a matrix of size k by m, where m is the number of genes (rows) in the dataset, and k the number of converged attractors. Each row of the matrix is the MI of all the genes with the converged attractor.

Missing values are not allowed as the input to the function in the current version.

Wei-Yi Cheng

Wei-Yi Cheng, Tai-Hsien Ou Yang and Dimitris Anastassiou, Biomolecular events in cancer revealed by attractor metagenes, PLoS Computational Biology, Vol. 9, Issue 2, February 2013.

findAttractor,parAttractorScanning, attractorScanningGL

# Load the toy dataset extracted from TCGA OV data
data(ov)

# Using the createFolds in the caret package to allocate tasks, assuming
# there are 100 workers
require(caret)
taskList <- createFolds(1:nrow(ov), k=100)

# Take the tasks given to worker 1
as <- parAttractorScanning(ov, taskList=taskList, wid=1)

# display the top 20 genes in first attractor
sort(as[1,], decreasing=TRUE)[1:20]