Developing computational methods for fundamentals and metrology of mass spectrometry imaging

[摘要] MSI is a suite of powerful imaging tools that can be used to perform untargeted unlabelled analysis into the distribution of a wide range of molecules from a variety of different sample types. Despite widespread use in numerous different research areas, many aspects of MSI fundamentals remain unknown. Not only are experimental aspects such as desorption and ionisation not always fully understood, but the success (or failure) of many of the computational methods used to mine these data cannot yet be easily evaluated. In this thesis, multivariate analysis methods are used to investigate fundamentals of laser parameters in raster mode MALDI imaging, and DF and CF variables in LESA coupled to FAIMS. Following this, novel methods to evaluate clustering algorithms are described, including multivariate normality testing for distance metric evaluation, and means to generate synthetic data based on multivariate normal distribution sampling. These synthetic data are then used to evaluate a variety of different clustering algorithms used previously in MSI and other fields, and a new, more efficient algorithm using graph based clustering and a two phase subset sampling approach is described. This is then demonstrated on large synthetic and real MSI datasets producing extremely accurate and informative segmentation.

[发布机构] University:University of Birmingham;Department:School of Chemistry

[关键词] T Technology;TK Electrical engineering. Electronics Nuclear engineering