A0566
Title: Statistical and machine learning models from resolving metabolomics spectra
Authors: Nikolaos Terzis - University of Glasgow (United Kingdom)
Andrew Elliott - University of Glasgow (United Kingdom)
Ronan Daly - University of Glasgow (United Kingdom)
Joe Wandy - University of Glasgow (United Kingdom)
Vinny Davies - University of Glasgow (United Kingdom) [presenting]
Abstract: Untargeted metabolomics experiments aim to identify the small molecules that make up a particular sample, e.g., blood or urine. The sample is put through a mass spectrometer, which performs multiple scans of different types, giving us a large amount of data which can be used to estimate the spectra which are needed to identify the metabolites. In a particular type of experiment, known as data-independent acquisition, large amounts of data representing metabolite fingerprints, known as fragments, are collected, which can be used to reconstruct the metabolite spectra within the samples. The challenge, however, with this method is that it is not clear to which metabolite a given fragment belongs. Data related to the abundance of the metabolites is used within the samples to link the measured fragments back to the metabolite from which they originated. A modelling framework is created that uses a large number of high dimensional regression models to create predictions for the metabolite spectra. The statistical framework is introduced, and a multitude of modelling methods used are described to tackle this difficult and very noisy task.
