Abstract
Genome-wide association studies have revealed that the genetic architecture of most complex traits is characterized by a large number of distinct effects scattered across the genome. Functional enrichment analyses help provide some biological interpretation of associated variants but more work is needed to further translate GWAS hits into meaningful biological insights. Thus, we set out to leverage the genetic association results from many traits with a view to identifying the set of modules, or latent factors, that mediate these associations. The identification of such modules may aid in disease classification as well as the elucidation of complex disease mechanisms. We propose a method, Genetic Unmixing by Independent Decomposition (GUIDE), to estimate a set of statistically independent latent factors that best express the patterns of association across many traits. The resulting latent factors not only have desirable mathematical properties, such as sparsity and a higher variance explained for the latent factors that are significantly associated with a given trait, but are also able to single out and prioritize key biological features or pathophysiological mechanisms underlying a given trait or disease. Moreover, we show that these latent factors can isolate biological pathways as well as epidemiological and environmental influences that compose the genetic architecture of complex traits.