A flexible R package for nonnegative matrix factorization

Abstract

BackgroundNonnegative Matrix Factorization (NMF) is an unsupervised learning technique that has been applied successfully in several fields, including signal processing, face recognition and text mining. Recent applications of NMF in bioinformatics have demonstrated its ability to extract meaningful information from high-dimensional data such as gene expression microarrays. Developments in NMF theory and applications have resulted in a variety of algorithms and methods. However, most NMF implementations have been on commercial platforms, while those that are freely available typically require programming skills. This limits their use by the wider research community.ResultsOur objective is to provide the bioinformatics community with an open-source, easy-to-use and unified interface to standard NMF algorithms, as well as with a simple framework to help implement and test new NMF methods. For that purpose, we have developed a package for the R/BioConductor platform. The package ports public code to R, and is structured to enable users to easily modify and/or add algorithms. It includes a number of published NMF algorithms and initialization methods and facilitates the combination of these to produce new NMF strategies. Commonly used benchmark data and visualization methods are provided to help in the comparison and interpretation of the results.ConclusionsThe NMF package helps realize the potential of Nonnegative Matrix Factorization, especially in bioinformatics, providing easy access to methods that have already yielded new insights in many applications. Documentation, source code and sample data are available from CRAN.

Evidence of Spreading Layer Emission in a Thermonuclear Superburst

“…We have shown that by using non-negative matrix factorization, the superburst spectra of 4U 1636-536 can be decomposed into two variable spectral components: the cooling burst spectrum, and a boundary/spreading layer, which is found to be contributing a sizable fraction of the total luminosity during the superburst. The spectral properties of the boundary/spreading layer component favors the spreading layer model (Inogamov & Sunyaev 1999, 2010Suleimanov & Poutanen 2006), where the spectrum is a constant ∼2.5 keV, quasi-Planckian component varying just in normalization as the burst evolves. This component is also very reminiscent of the frequency-resolved spectral component of a constant spectral shape that is responsible for the sub-second variability in many NSLMXBs.…”

“…For calculating the decomposition we used the package nmf (Gaujoux & Seoighe 2010) that calculates the standard NMF (Brunet et al 2004) by picking random starting values for W and S from a uniform distribution [0,max(X)] and then updating iteratively 10000 times to find a local minimum of the cost function with a multiplicative rule from Lee & Seung (2001). The minimization process is repeated for 300 starting points to ensure that the algorithm does not get stuck in a local minimum.…”

“…As the source moves from the hard state to the soft ("banana") spectral state, the boundary layer becomes optically thick to Comptonization and cools down causing a rapid drop in the X-ray hardness ratio (Done et al 2007). Two distinct geometries of the boundary layer are considered possible: In a classical boundary layer model (hereafter referred to as BL) the rotational velocity decreases over a large radial extent in the disc midplane (e.g., Pringle 1977;Sunyaev & Shakura 1986; 1 email: karri.koljonen@nyu.edu Popham & Sunyaev 2001), or in the alternative spreading layer model (hereafter referred to as SL) the radial extent of the layer is smaller but the gas spreads over a considerable height from the equatorial plane towards higher stellar latitudes (e.g., Inogamov & Sunyaev 1999, 2010. Radiative transfer calculations of the BL model by Grebenev & Sunyaev (2002) predict spectra that are harder than what is observed in NS-LMXBs in the soft spectral state, whereas the SL model calculations by Suleimanov & Poutanen (2006) predict softer, quasiPlanckian spectra with a color temperature of 2.4-2.6 keV as long as the accretion rate (Ṁ ) is sufficiently high (> 10 % of the Eddington accretion rate).…”

Low immunoglobulin E flags two distinct types of immune dysregulation

“…Heatmaps visualizing the hierarchical clustering of immunoglobulin profiles were established using the hclust-R function with default parameters and plotted using the heatmap function from the NMF R package [14].…”

A non-negative matrix factorization framework for identifying modular patterns in metagenomic profile data

“…The NMF package (Gaujoux and Seoighe 2010) picks random starting values for W and H and then updates iteratively to find a local minimum of KL divergence. We repeated this process for 100 different starting points for each value of k, and used the results to evaluate the quality and stability of the factorization at this degree.…”