LS-NMF: A modified non-negative matrix factorization algorithm utilizing uncertainty estimates-2

<p><b>Copyright information:</b></p><p>Taken from "LS-NMF: A modified non-negative matrix factorization algorithm utilizing uncertainty estimates"</p><p>BMC Bioinformatics 2006;7():175-175.</p><p>Published online 28 Mar 2006</p><p>PMCID:PMC1450309.</p><p>Copyright © 2006 Wang et al; licensee BioMed Central Ltd.</p>, while section c shows the results for the Rosetta Compendium analysis, where variance estimates are available. In a, the total area under the ROC curves is shown for LS-NMF (solid line) and NMF (dashed line). In b, LS-NMF and NMF are compared to hierarchical clustering at a single dimensionality (the LS-NMF and NMF curves are superimposed as there is no difference). In c, the areas under the ROC curve for analysis of the Rosetta data are shown for LS-NMF (solid line), NMF (dash-dot line), K-means clustering (dashed line), and NMF on scaled data (C-NMF, dotted line). Here the variance information allows far better results to be obtained, either by scaling or by use of LS-NMF. The advantages of the approach used in LS-NMF are discussed in the text