Wednesday, November 01, 2017 4:00pm - 5:00pm
Watson CIT - SWIG Boardroom (CIT241)
Dept of Biological Statistics and Computational Biology
Leveraging and inferring properties of ungenotyped ancestors enables high resolution relatedness detection
Pedigree relationships are fundamental to genetics, with relatedness estimation needed to perform nearly all genetic analyses from trait association mapping to demographic reconstruction. Until recently, methods for inferring relatedness considered each pair of samples independently even though multiple samples have the potential to be mutually related. We have developed DRUID—Deep Relatedness Utilizing Identity by Descent—a method that works by inferring the IBD sharing profile of an ungenotyped ancestor of a set of close relatives and then inferring relatedness between that ancestor and a distant relative. We used DRUID to infer relatedness in real data and demonstrate that it classifies up to 24% more samples correctly compared to traditional pairwise approaches, and 6% more samples compared to other multi-way relatedness methods. Building off these insights, we describe a novel approach for inferring the partial genomes of the parents of a set of siblings. Despite ambiguity corresponding to >4 million possible assignments of autosomal haplotypes to the parents, differences in male and female crossover patterns enable very confident inferences of the correct assignment. This method presently applies when data for large numbers of siblings are available, and we sketch extensions of it to apply to data for smaller numbers of siblings.