engleski

COMPUTATIONALLY EFFICINET SEPARATION OF LARGE NUMBER OF ANALYTES FROM SMALL NUMBER OF MIXTURE MASS SPECTRA

Contemporary metabolic profiling of biological samples aims to extract pure components from possibly nonlinear multicomponent mixtures mass spectra. Due to large number of components present in the mixtures spectra, related separation problem is hard. Herein, we further elaborate reproducible kernel Hilbert space method for nonlinear underdetermined blind source separation (uBSS) of components from mixtures mass spectra. In particular, we compare accuracy of approximate, but computationally more efficient method with the original one, developed recently on two problems related to: extraction of 25 components form 9 mixtures mass spectra, and extraction of 19 components from 12 mixtures mass spectra. Mixtures mass spectra (m1-mn) were obtained from chemical reaction of peptide formation at different time points (t0-tm-1). By using approximate version of pure components extraction method it is possible to drastically reduce computation time while still retaining decent separation quality. This result is of practical relevance, because approximate method combined with high-speed implementation on graphical processing units enables nearly real time execution. Here presented approach is of wide application, such as non-targeted metabolite profiling or dynamic libraries search.

pure components; mass spectrometry; nonlinear blind source separation

nije evidentirano