COMPADRE: Combined pedigree-aware distant relatedness estimation for improved pedigree reconstruction

Designing powerful and unbiased genomic studies requires accurate assessment of familial relatedness even when this information is not captured from participants. Characterization of pairwise degrees of relatedness from participants' genetic data enables reconstruction of pedigrees, and several pedigree reconstruction tools have emerged in the last decade. However, limitations of these tools include high computational burden in large datasets, reliance on external information, reduced accuracy in admixed populations, and most notably, an inability to accurately reconstruct pedigrees when only a subset of family members is represented in the genetic data. To improve pedigree reconstruction in large-scale data and in pedigrees with high proportions of ungenotyped individuals, we have developed a tool, combined pedigree-aware distant relatedness estimation (COMPADRE), that leverages pairwise relationship estimates informed by both averaged genome-wide identity-by-descent (IBD) sharing estimates and shared IBD segments' length and distribution. COMPADRE reduces the number of candidate pedigrees compared to pedigree reconstruction and identification of a maximum unrelated set (PRIMUS), resulting in lower resource utilization and fewer high-likelihood pedigrees reconstructed, with the greatest improvements observed in pedigrees with high sample missingness. In 20- and 40-person pedigree simulations, COMPADRE also increases the proportions of correctly reconstructed pedigrees compared to PRIMUS, Bonsai, and CLAPPER. In 15,478 individuals of African ancestry from the BioVU biobank, COMPADRE achieved greater precision relative to PRIMUS. COMPADRE represents a marked improvement in both the confidence and accuracy of pedigree reconstruction relative to existing tools. COMPADRE integrates updates to the PRIMUS algorithm with the functions of estimation of recent shared ancestry (ERSA) and PADRE within a single tool and is freely available online.