Developing statistical learning methods to infer how higher-order genomic architecture influences phenotypes (360G-Wellcome-203950_Z_16_A)

Since the completion of the Human Genome Project, big genomic data is becoming available at an unprecedented scale. The early genome-wide association studies (GWAS) that followed, which sought to relate phenotypic differences to genomic variation, revealed that many diseases have complex aetiologies, where a combination of genetic and environmental factors contribute to the development of traits, with their relative importance quantified by heritability. Simultaneously, in recent years, neural-networks emerged as a powerful framework to solve non-linear prediction tasks, but hitherto have not been applied to the GWAS problem yet. During my first year, I developed an artificial deep neural-network prototype, which utilised heritability to obtain a higher phenotype prediction accuracy, compared to baseline models. In my full PhD project, I will build on my early successes. First, I will make my framework more applicable to larger and smaller datasets. Then I will move onto creating tools that may provide insight into the aetiology of traits and finally, I plan to expand my methods to incorporate multi-omic data into the model. I hope that my research project will contribute to the development of tailored therapies that may provide a more effective approach to prevent or treat complex diseases in the future.