Diego Oyarzun, d.oyarzun@ed.ac.uk
Oisin Mac Aodha, oisin.macaodha@ed.ac.uk
Project Description
The project aims to improve the analysis of cellular metabolism for therapy and diagnostics, using genome-scale metabolic models (GEMs) and self-supervised learning. Cellular metabolism is a complex network of thousands of components, which makes it challenging to predict systemic effects when specific reactions are inhibited. Genome-scale metabolic models offer a mathematical framework for describing the connectivity of metabolic fluxes, but their high dimensionality limits our ability to draw effective predictions. We aim to develop an end-to-end pipeline for graph representation learning of GEMs, test the utility of low-dimensional embeddings for therapy-relevant prediction tasks, and build a user-friendly web platform for model training. The project will rely on well-adopted packages for GEM analysis and simulation and generate fluxomic data for model training. We will employ a combination of variational autoencoders and metabolic graphs to learn low-dimensional embeddings of the metabolic space. The project will deliver a suite of data, models, and software for analysing metabolism in disease, potentially contributing to the discovery of therapeutic targets or diagnostic biomarkers across various conditions.
