Machine Learning of Genetic, Clinical and Environmental Data for Early Morbidity Detection in the UK Biobank.

Niels Bohr stated that 'It is Difficult to Make Predictions, Especially About the Future' but the development of data science methods allows us to build increasingly effective predictive models in large data sets. This PhD project will apply machine learning and deep learning methods, as well as classic statistical models, to the UK Biobank, an incredible health study of over 500,000 people in the UK. The student will integrate genetic, environmental and clinical data to predict onset of diseases that are relevant for the UK's aging population such as heart disease and cancer. A particular focus will be assessing the utility of genetic information: does genetics add information to routinely-collected clinical and biomarker data, and what role could genetics play in clinical prediction algorithms?

In Year 1, the student will develop their programming, analytical and 'big data' skills, building classic statistical models and machine learning algorithms to assess the predictive ability of clinical data (including biometrics and blood biomarkers), lifestyle data (such as smoking habits, diet and exercise) and genetic predisposition in coronary artery disease. In Year 2, novel genetic risk scores will be built for different disorders, using machine learning methods, and their predictive ability assessed, in combination with all other sources of information. In addition, machine/deep learning methods will be used to identify new environmental risk factors. In Year 3, the student will build comprehensive disease risk models and test their predictive power against the gold-standard clinical prediction tools.