Pangenome-Wide Association Studies with Frequented Regions

Connecting genetic variation (genotype) to trait variation (phenotype) is a critical but often difficult step in genetic research. A genome-wide association study (GWAS) is a common approach to connect underlying genetic variation to complex phenotypic traits, allowing for phenotypic prediction. GWAS is important in many disciplines, including identifying genetic risk factors for common, complex diseases, identifying genes underlying important traits and predicting phenotypes from genotypes. GWAS is limited, though, in that the types of variations typically studied are single nucleotide polymorphisms (SNPs) identified relative to a single reference genome. These limitations lead to bias and preclude GWAS from studies across related species. The advent of next-generation sequencing has brought an exponential growth in DNA sequence data. This has led to the more comprehensive pangenomics approach, where the entire sequence content and variation of a population are succinctly represented independent of a reference. In prior work, we developed a method for identifying genomic regions that characterize complex variations within pangenomic data and showed that these regions provide a more general way to study genetic variation than existing approaches. This work describes our initial results to develop new methods for a new branch of genomic analysis called pangenome-wide association studies (PWAS) that generalizes GWAS to pangenome datasets both within and across species. We make use of recently developed algorithms for fast compressed De Bruijn graph construction and identifying frequented regions in these graphs that can be used as machine-learning features to identify pangenomic regions, overlaid with gene annotations, that relate to complex phenotypic traits. Initial results on a pangenome composed of 100 yeast indicate that frequented region features provide better machine-learning regression models than SNPs for predicting phenotypic traits.