Pathway-driven integration and interpretation of metabolomics and multi-omics data

Abstract

Pathway analysis is a widespread bioinformatics method for functional interpretation of omics data. Despite its popularity, there is little research, standardisation, or benchmarking of its applicability tometabolomics data,which is often neglected in favour of sequencingbased omics. Metabolomics,which involves the high-throughput profiling of small molecules, is becoming increasingly popular, especially in conjunction with other omics, underscoring an additional need for interpretable, user-friendly multi-omics integration methods. In this thesis I have focused on evaluating and developing methods for metabolomics pathway analysis and pathway-based multi-omics integration. First, I investigated the suitability of the popular over-representation pathway analysis approach for metabolomics data, systematically evaluating a series of essential input parameters andmass-spectrometry data-specific considerations, and suggesting best practice recommendations. Second, I evaluated for the first time the suitability of single-sample pathway analysis for metabolomics data, demonstrating the performance of established and newly proposed methods on both simulated and experimental datasets. Lastly, I leveraged single-sample pathway analysis to transformmolecular-level omics data to the pathway dimension, alongside state-of-the-art multivariate predictive models to create the PathIntegrate framework for pathway-based multi-omics integration. PathIntegrate combines multi-omics data at the pathway level, able to accurately identify pathway perturbations and predict sample outcomes even in low signal-to-noise scenarios, and importantly provides a set of readily-interpretable outputs in the formof multi-omics pathways. Taken together, this work builds towards standardising metabolomics pathway analysis workflows, raising awareness of the limitations and data-specific considerations, whilst proposing innovative methods for tackling biological interpretation and integration.