Hierarchical Biclustering of Mouse Pancreas Mass Spectrometry\nImaging Data Using Recursive Rank‑2 Non-negative Matrix Factorization

One\nof the main challenges in mass spectrometry imaging data analysis\nremains the analysis of <i>m</i>/<i>z</i>-spectra\ndisplaying a low signal-to-noise ratio caused by their low abundance,\nsample preparation, matrix effects, fragmentation, and other artifacts.\nAdditionally, we observe that molecules with a high abundance suppress\nthose with lower intensities and misdirect classical tools for MSI\ndata analysis, such as principal component analysis. As a result,\nthe observed significance of a molecule may not always be directly\nrelated to its abundance. In this work, we present a recursive rank-2\nnon-negative matrix factorization (rr2-NMF) algorithm that automatically\nreturns spectral and spatial visualization of colocalized molecules,\nboth highly and lowly abundant. Using this hierarchical decomposition,\nour method finds spatial and spectral correlations on different levels\nof abundances. The quality of the analysis is evaluated on MALDI-TOF\ndata of healthy mouse pancreatic tissue for the annotation of molecules\nof interest in the lower abundances. The results show interesting\nfindings regarding the functioning and colocalization of certain molecules.