Classification of pyrolysis mass spectra of biological materials using convex cones

This work addresses the classification of high-dimensional time-dependent pyrolysis mass spectra of biological samples. The aim was the detection and classification of biological agents, and the developed approach resembles mixture analysis. The data were projected on to a low-dimensional subspace using singular value decomposition. Then a convex cone was formed on this subspace, showing as its corners physically meaningful components of the sample. This technique enabled separation of a biological material signal largely independent of the absolute amount of sample. The detection of the presence of any biological material could be accomplished based on the convex cone alone, without other reference to the mass spectra. Automated clustering of samples was successfully carried out using a minimal spanning tree.