bicluster-methods: Biclustering algorithms
In jonalim/mfBiclust: Pipelines for Bi-Clustering Using Matrix Factorization

Description Details See Also

The method argument of addStrat() provides access to internal mfBiclust functions, which themselves call functions from packages NMF and biclust as back-ends. Back-end arguments for "als-nmf", "svd-pca" and "nipals-pca" are described on their respective pages. For arguments specific to other methods, please see respective documentation in other packages.

als_nmf: Alternating-least-squares non-negative matrix approximation. Fast at low values of k, but rapidly slows as k increases.
svd_pca: The Singular value decmoposition algorithm. Each principal component is interpreted as the degree of membership in a single bicluster. The resulting score matrix is thresholded to binarize bicluster membership. svd_pca is the fastest provided algorithm.
nipals_pca: An iterative PCA algorithm that may tolerate missing data. Slower than svd_pca, but still faster than the other algorithms.
plaid: Uses biclust::biclust(method = BCPlaid(), ...) as its back-end. To get k biclusters, back-end arguments row.release and col.release are simultaneously decremented towards 0.1 in steps of 0.1, then shuffle is incremented towards 10 in steps of 1.
snmf: Sparse non-negative matrix factorization. Uses NMF::nmf(method = "snmf/r", ...) as its back-end. To get k biclusters, back-end arguments eta and beta are initialized at mean(A) and are halved progressively.
spectral: Uses biclust::biclust(method = BCSpectral(), ...) as its back-end. For smaller matrices, the back-end argument numberOfEigenvalues is automatically set to enable finding the number of biclusters requested. The back-end argument withinVar is initialized equal to the smaller matrix dimension and allowed to increase up to 10 times the smaller matrix dimension to find k biclusters.