Bayesian Deconvolution and Quantitation of NMR Metabolic Profiles with BATMAN
Timothy Ebbels (Imperial College London)
Tuesday 8th November, 2016 15:00-16:00 Sir Alwyn Williams Building, Level 5
Metabolomics is the study of metabolites – the small molecules which provide the energy and building blocks of life. Levels of metabolites in biofluids and tissues are a sensitive barometer of biological status, and can reveal a great deal about the health and disease. These levels are often measured using Nuclear Magnetic Resonance (NMR) spectroscopy which produces complex spectra, typically consisting of hundreds of overlapped peaks. Peak annotation, deconvolution and quantitation are key bottlenecks in the current metabolomics pipeline and thus key targets for new computational approaches. Current solutions are problematic in that they either require intensive manual effort, or simply defer these problems to later stages of the workflow.
To tackle this challenge, we developed BATMAN – the Bayesian AuTomated Metabolite Analyser for NMR spectra which decomposes spectra of complex mixtures into their constituent pure components. For known metabolites, BATMAN uses resonance patterns from the Human Metabolome Database. This extensive prior information also allows us to deconvolve overlapped resonances and obtain concentration estimates for the corresponding metabolites. Spectral signals without prior information are modelled using wavelets. In performance tests, BATMAN enabled semi-automated quantification of metabolites from biofluids with similar accuracy to current manual approaches, thus helping to streamline the global NMR metabolomics workflow and accelerate metabolic knowledge discovery.