Piotr Dworszynski - PREDICTING T2D USING GENETIC RISK SCORES AND LARGE-SCALE PHENOTYPIC DATA

As emergence of precision medicine brings along the great promise of health-care decisions tailored to the individual, the discovery of early risk factors becomes paramount. The growth of longitudinal biomedical data including nation-wide registers, genetics, health records and metabolomics creates a unique opportunity to develop novel methods for predicting of onset, trajectories and outcomes of complex diseases early and in a non-invasive manner. Leveraging these diverse datasets requires application of state-of-the-art machine learning methods able to capture the complex relationships between variables and finding new ways in which missing data features can be addressed without loss of information.

These challenges can be addressed by integrating genetics data, world knowledge, network similarity profiles as well as novel temporal data representations. Studying these methods raises hope of uncovering novel type 2 diabetes (T2D) biomarkers which may allow for early identification of at-risk individuals and through it personalized interventions preventing, rather than delaying, disease onset.

My overarching aim is thus to employ semantic knowledge to integrate high-dimensional, temporal biomedical data sources with machine learning to improve prediction of T2D and other biomedical outcomes. I will do so by first, constructing genetic risk scores based on causal intermediate phenotypes for T2D and other complex diseases. Second, by leveraging knowledge graph representations of biomedical data to coalesce distinct events in biomedical datasets and apply advanced machine learning approaches to predict complex disease outcomes. Lastly, by performing network-based prediction of biomedical outcomes by leveraging similarity profiles between individuals based on genetic risk scores, questionnaire data, physiological data, registry-based information and laboratory measurements.