Plant seed classification using pyrolysis mass spectrometry with unsupervised learning: The application of auto-associative and Kohonen artificial neural networks

摘要: Abstract Pyrolysis mass spectrometry (PyMS) was used to gain high dimensional (150 m/z values) biochemical fingerprints from Begonia semperflorens Summer Rainbow, Campanula carpatica White Gem, Lobelia erinus Fountain, and Lady plant seeds. Rather than homogenizing the seeds analysing extracts, sample preparation of in this study novel merely involved crimping metal foil carrier around Compared extractive procedures technique exploited will give a fair representation seed, is rapid thus amenable analysis volume samples. To observe relationship between these seeds, based on their spectral fingerprints, it necessary reduce dimensionality data by unsupervised feature extraction methods. The neural computational pattern recognition techniques self organising maps (SOFMs) auto-associative networks were therefore employed clusters observed compared with groups obtained more conventional statistical approaches principal components (PCA) canonical variates (CVA). When PCA analyze raw pyrolysis spectra replicate samples not recovered discrete clusters; CVA, which minimises within-group variance maximises between-group variance, had be employed. Although B. C. separately away L. , two types could still discriminated using approach. CVA uses priori information are replicates; we encoded employing preprocessing regime where triplicate each averaged pairs produce three new spectra; then failed separate ; however, successfully discriminate them. It likely that due ability perform non-linear mappings hence approximate PCA. SOFMs also all four unequivocally. obtain quantitative regarding similarity spectra, trained different numbers nodes Kohonen output layers. results procedure often difficult report tables or visualise topological contour maps; simplify graphical performed construction dendrogram various analyses. This demonstrates potential PyMS for discriminating at genus, species sub-species level. Moreover true reflection known taxonomy plants. approach invaluable taxonomist representing biological relationships among taxa describing genomic without need cultivation propagule.