Using statistics and machine learning to create a new metabolomics fragmentation spectra resolver

Untargeted metabolomics experiments aim to identify the small molecules that make up a particular sample, e.g., blood, urine. The sample is put through a mass spectrometer which scans the sample in multiple ways to help us work out what metabolites can be found. Identifying these metabolites can be useful for clinical trials, disease diagnosis and progression, and various other applications. There are various ways of choosing the scans in order to optimally collect information about the sample, but in one particular method, DIA, we often see ion fragments from multiple metabolites in a single scan. In order to assign individual fragmentation spectra to metabolites within the sample we must work out which of the fragments in each of the scans belongs to each observed metabolite. This project will make use of a recently created virtual mass spectrometer, ViMMS, combined with data analytics methods from machine learning and statistics, e.g., LASSO, random forest, to predict the relationship between the scans and the metabolites. The methods will be evaluated, and data extraction methods optimised in order to test whether these methods can outperform current state of the art methods. Over the project, the method will be extended to deal with case of multiple identical and then different samples