A MACHINE LEARNING APPROACH TO UNDERSTAND HOW INVERTEBRATES MAINTAIN HOMEOSTASIS UPON EXPOSURE TO PHARMACEUTICALS.

The aim of this project will be to understand fundamental metabolomic and lipidomic homeostatic mechanisms in a freshwater invertebrate using machine learning to interrogate relationships.
Objectives:
1. To develop untargeted and targeted analytical (LC-MS/MS and GC-MS) methods for the determination of various classes of lipids (sphingolipids, phospholipids, triglycerides, fatty acids, etc.) from samples already collected and stored from BBSRC IPA project iNVERTOX (Ref: BB/P005187/1). A method for untargeted analysis will also be developed using full scan GC-MS and LC high resolution mass spectrometry (using hydrophilic interaction and reversed-phase liquid chromatography).
2. To establish a baseline lipid profile of the selected species (Gammarus pulex) and to use in silico classification tools such as principal component analysis and/or partial least squares analysis to identify confounding factors that influence the variability (e.g., sex, age, moult cycle and acclimatisation period).
3. Use convolutional neural networks for classification of LC- and GC-MS ion maps determined from Obj. 1 (lipidome) and iNVERTOX (metabolome) to enable classification and fundamental interpretation of cross-omic techniques.
4. Exposure of G. pulex to 5-10 selected pharmaceuticals at several concentrations to determine dose-response curves (internal concentrations) that disrupt homeostatic processes. Metabolomics and lipidomics will also be performed on these samples.
5. To develop and apply machine learning tools to the metabolomics and lipidomics data gathered in Obj. 4 to identify metabolic features across both datasets that are linked and correlated with dose-response curves and potential pathways. The metabolism of the probe pharmaceutical compounds will also be performed (via reversed-phase LC-HRMS)