Towards precision medicine : multimodal machine learning approaches for personalised risk prediction in cancer, Parkinson's diseases, and chronic ageing-related diseases

Abstract

Precision medicine, or personalised medicine, focuses on customising disease prevention and treatment for individual patients based on their unique genetic features, environmental factors, and lifestyle. It encompasses a range of crucial tasks, such as disease prevention, early diagnosis, personalised treatment, and the risk prediction of disease progression and mortality. These personalised health-related risk tasks rely heavily on identifying meaningful biomarkers to support accurate predictions. Previous studies have shown the importance of risk biomarkers in various disease tasks. Despite its potential, precision medicine faces significant challenges due to the diverse and heterogeneous nature of patient profiles. To comprehensively understand and address the complexities of personalised patient care and propose meaningful personalised disease risk biomarkers, multimodal data from various sources is required. This thesis aims to solve the challenge of predicting health risks by proposing and applying machine learning (ML) models that can easily combine different sources of data and develop personalised disease risk biomarkers for more accurate predictions. The focus is on three fundamental health-related risks: mortality risk, prognostic risk, and disease diagnostic risk. To achieve this, various data modalities are included, such as laboratory blood samples, radiology images, genetic data, clinical records, questionnaires, and more. Those studies place significant emphasis on the prediction of risks, including cancer, multi-ageing related disorders, and Parkinson's disease (PD). Specialised ML models are implemented for each risk prediction task to advance personalised healthcare through the effective utilisation of multimodal data. By incorporating various data modalities and employing state-of-the-art ML methods, this thesis aims to improve the precision of health-related risk prediction and contribute to the wider domain of precision medicine.