Project Summary This project aims to advance our understanding of major depressive disorder (MDD) through the analysis of electronic medical records, biobanks and associated genetic data. MDD is the commonest psychiatric disorder and recognized as the world’s leading cause of disability, yet current treatments are relatively ineffective: only about half of patients will show signs of improvement after three months of therapy. Genetic approaches are a proven path to identifying causal factors and hence finding novel treatments, but they are hard to apply to MDD without obtaining large samples of cases. We propose using the very large numbers of cases available through electronic medical records by applying statistical methods that accurately identify MDD. Our methods provide a “best-guess” diagnosis by a process known as imputation. We then identify features that are specific to MDD. Our insight is that since non-genetic and non-specific factors explain large components of variability in traditional MDD phenotypes, algorithmically removing them increases the signal from the core biological drivers. We assume that non-specificity can be attributed to latent factors capturing the relationship between MDD, comorbid disease, and pleiotropic factors. By identifying and removing these signals, we increase specificity, and thus identify features that reflect the episodic severe shifts of mood, associated with neurovegetative and cognitive changes, that are central to MDD. Our project has three aims: first, to impute phenotypes of a large sample of MDD cases and controls in biobank data and determine the best approximation to MDD; second, to identify and characterise specific and non-specific genetic effects on MDD, and finally to identify genes involved in MDD by associating the cases defined via our first two aims with rare coding variants.