Annotating metabolites with topic models
Simon Rogers (University of Glasgow)
Friday 6th May, 2016 15:00-16:00 Maths 522
Analysis of high throughput metabolomics experiments is currently hampered by our inability to identify the molecules from the peaks they produce in mass spectra. Fragmentation of molecules is used to partially overcome this problem but still leaves many metabolites un-identified. In this talk I will describe an approach to overcome this problem that uses topic modelling techniques developed for text analysis to decompose fragment spectra into small collections of fragments (and losses) indicative of common chemical substructures. Manual annotation of these substructures (a reasonably straightforward task) allows many more metabolites to be meaningfully annotated than more traditional methods.