Representation learning for biological networks and multi-omics integration

PhD defence by Dewei Hu

Assessment Committee

Associate professor Jessica Xin Hjaltelin, Department of Public Health, University of Copenhagen (Chairperson)
Associate professor Alberto Santos Delgado, Technical University of Denmark
Associate professor Pedro Beltrao, Federal Institute of Technology Zurich (ETH Zurich)

Supervisors

Professor Simon Rasmussen
Professor Lars Juhl Jensen

Department

Novo Nordisk Foundation Center for Basic Metabolic Research

Graduate Programme

Biostatistics and Bioinformatics

Place

Panum. Building 13, Room: Area in front of Jerne auditorium, Blegdamsvej 3B, 2200 København

Email address to gain access to the thesis: dewei.hu@sund.ku.dk.
You will either receive a copy of the thesis or be informed where you can read a physical copy.
Recipients of copies of the thesis are not allowed to share or distribute it due to copyright compliance.

Short description of the thesis

Proteins run almost every process in our cells, working together in vast networks, yet what many of them do, and how they fail in disease, remains unknown. This thesis uses machine learning to distil these networks and other molecular data into compact numerical profiles that computers can learn from. These profiles help predict the roles of unstudied proteins across over a thousand species, uncover two to four times as many disease-relevant proteins, and power a simple model that forecasts how cells respond when genes are switched off, rivalling far larger AI models. Finally, aligning profiles of costly protein measurements with those of cheap blood tests improves risk prediction for 82 diseases, bringing earlier, more personalised prevention a step closer.