Uncovering UK Biobank Lifestyle Patterns with Deep Learning

UK Biobank is a large biomedical database, providing both phenotypical and genetical information about roughly 500,000 individuals living in the United Kingdom. Lifestyle information is of a particular interest, having been the target of several studies over the years, trying to link lifestyle choices and food intake to health outcomes. However, large data availability comes with high complexity, in terms of missing data, unprovided responses to surveys and complex interactions among different factors, that are not so easily predictable and yet may be much informative. A novel approach is here suggested, leveraging Deep Learning techniques. Different types of neural networks were trained on a set of “environmental” variables, summarizing life habits and dietary choices, and anthropometric measurements such as BMI. Leveraging supervised and unsupervised methods, Transformers-based architectures has shown to be effective in providing an informative representation of the manifold space of “lifestyle variables”.