Public defence in Computer Science, M.Sc. Sophie Wharrie

Title of the thesis: Advancing Towards Personalized Medicine: Probabilistic Machine Learning and Deep Learning for Health and Genetics

Thesis defender: Sophie Wharrie
Opponent: Professor Hanna Suominen, Australian National University
Custos: Professor Samuel Kaski, Aalto University School of Science

Personalized medicine aims to meet the unique diagnosis, prevention, and treatment needs of individuals by considering factors such as genetics and medical history. Artificial intelligence and machine learning can facilitate a data-driven approach to personalized medicine by leveraging vast amounts of complex health and genetics data. However, we only observe a limited number of data points per person and disease profiles differ significantly across individuals, posing challenges for machine learning for “personalized” medicine. The research presented in the thesis seeks to develop new machine learning techniques to address these challenges and was conducted using real-world data from large health registries and biobanks in Finland and the United Kingdom.

The research makes contributions to two areas in machine learning for personalized medicine. Firstly, it introduces a new probabilistic machine learning approach for generating high-quality synthetic data for genetics and disease phenotypes at the individual-level, which was applied to personalized models of genetic factors of disease. Second, it presents advanced deep learning techniques for more effectively utilising large data sources when predicting and explaining health outcomes from longitudinal data. This includes a geometric deep learning approach that leverages biological relationships between individuals, and a Bayesian meta-learning approach that considers causal relationships underlying the outcomes being predicted.

The research findings have significant potential for applications in personalized medicine, especially in improving how machine learning is applied to large-scale health and genetics data.

Thesis available for public display 10 days prior to the defence at Aaltodoc.