Application of non-negative matrix factorization to LC/MS data

Liquid Chromatography-Mass Spectrometry (LC/MS) provides large datasets from which one needs to extract the relevant information. Since these data are made of non-negative mixtures of non-negative mass spectra, non-negative matrix factorization (NMF) is well suited for their processing. These data are however very difficult to deal with since they are usually contaminated with non-Gaussian noise and the intensities vary on several orders of magnitude. In this paper, we propose an adaptation of a state-of-the-art NMF algorithms so as to specifically be able to deal with LC/MS data, by using a non-stationary noise model and a stochastic term. We finally perform experiments and compare standard NMF algorithms on both simulated data and an annotated LC/MS dataset. The results of these experiments highlight the significant improvement obtained by our adaptation over other NMF algorithms.