Machine learning methods for the automated and reproducible analysis of multi-parameter techniques in cytometry

Abstract

The human immune system is a complex network of organs, tissues and cells that work together to defend the body from invading pathogens. Our response to disease is mediated through specialised immune cells that are characterised by similar properties such as RNA expression or protein abundance. Cytometry is the field of single-cell proteomics that has revolutionised the study of expression, states, and types of cells. Computational immunology is the study of cells from technologies such as cytometry using computationally data-driven algorithms. Two broad subsets of these algorithms are supervised and unsupervised methods that have facilitated the discovery of novel cell types and unique immunophenotypic profiles that underpin disease. With an unprecedented rise in data deposition and accessibility, advances in these algorithms are required to study combined data sets at speed, scale, and resolution. In this thesis, I present Immunopred, a novel supervised deep learning-based tool used to identify immune cells. Immunopred is a robust approach that allows scientists to study cells from multiple studies in a combined way. I generated an extrinsic reference for healthy people comprising seventeen studies to support Immunopred's operation in any blood-based mass cytometry data set. Immunopred is applied to automatically identify 13 immune cell populations using only ten lineage markers. I critically evaluate Immunopred both quantitatively and using biomedically validated literature; including healthy control and COVID disease data sets, and recommendations, to demonstrate the successful applicability of the tool. I extended Immunopred to develop an end-to- end meta-analysis pipeline and test its applicability on other multimodal cytometry-based technologies. The impact of this research is to facilitate novel avenues into the large-scale integrative analysis of publicly available cytometry data. This will allow for insight into the pathogenesis of the disease by studying the functional roles of cells and antibodies. Ultimately, contributing to the discovery of drug candidates and life-changing clinical applications.