Discovering genetic drivers of longitudinal electronic health records (360G-Wellcome-215096_Z_18_Z)

The accumulation of electronic health records (EHRs) carries great promise for accelerating clinical research, while the sequencing of a vast number of genomes provides us information to discover genetic factors affecting these health-related records. We now have EHRs containing phenotypes and the genetics of individuals which we could use find the underlying genetic structure of disease. In our project, we will be focusing on analysing the development of disease using multiple machine learning methods. We propose a multi-view deep learning framework which incorporates embedding learning, recurrent neural networks, representation learning and weighted-averaging techniques. Our goal is to find the genetic factors which impact the progression of disease, not just the onset of disease. Consequently, our model will look into the time dimension of disease using multimodal data in huge datasets like the UK Biobank. Our results will contribute to our understanding of underlying systems and processes that can impact multiple diseases, which can support clinical diagnosis and drug discovery.