bicBinCluster: Biclustering sparse binary data in parallel

In TylerBackman/BicBin: Biclustering Sparse Binary Data In Parallel

Description

Biclusters sparse binary data in parallel across many cores based on BicBin by Miranda van Uitert, Wouter Meuleman, and Lodewyk Wessels.

Usage

bicBinCluster(binaryMatrix, totalClusters, alpha = 0.5, beta = 0.5, minDensityP1 = 0, tries = 900)

Arguments

binaryMatrix	This is a binary matrix (0s and 1s).
totalClusters	The desired number of clusters to find.
alpha	The alpha term from the BicBin algorithm indicating how rows should be weighted (higher = more rows in resulting matrix). Raising both alpha and beta equally will find smaller and denser biclusters.
beta	The beta term from the BicBin algorithm indicating how columns should be weighted (higher = more columns in resulting matrix). Raising both alpha and beta equally will find smaller and denser biclusters.
minDensityP1	The minimum bicluster density (P_1 in algorithm A_2). Biclustering is run iteratively until a cluster of this density is found. Setting it to 1 will find non-sparse clusters, and setting it to 0 will return the best scoring clusters regardless of density.
tries	The number of random iterations to try. Users should try multiple values to test for convergence (i.e. the number needed to produce repeatable results).

Value

Returns a list of the bicluster coordinates and scores. The elements rows and cols contain the row and column numbers for the bicluster, as well as the score for that bicluster. See the below example for syntax on accessing rows, columns, and scores. Note that the scores may not mean much for comparing across biclusters, as they are calculated with a local probability p for the portion of the sub-matrix being considered at the time. Implementing the code in this way was necessary, or else the minDensityP1 option would not be possible, as one wouldn't get denser clusters on each re-evaluation.

Author(s)

Tyler William H Backman

References

Biclustering sparse binary genomic data. van Uitert, Meuleman W, Wessels L. J Comput Biol. 2008 Dec;15(10):1329-45. doi: 10.1089/cmb.2008.0066.

Examples

cores <- 1 # set cpu cores
gc() # clear unused memory
set.seed(123) # get same clusters each time
registerDoMC(cores=cores) # register parallel backend
tries <- cores*10 # make tries an even multiple of cores
mat <- matrix(sample(c(rep.int(0, 100), 1), 800*250, replace=TRUE), 800, 250)
bicluster <- bicBinCluster(binaryMatrix=mat, totalClusters=1,
	alpha=0.5, beta=0.8, tries=tries, minDensityP1=0)
firstBicluster <- mat[bicluster[[1]]$rows, bicluster[[1]]$cols]
firstBiclusterScore <- bicluster[[1]]$score

TylerBackman/BicBin documentation built on May 9, 2019, 5:17 p.m.