Improving Forensic Identification Using Bayesian Networks and Relatedness Estimation: Allowing for Population Substructure

Abstract

Population substructure refers to any population that does not randomly mate. In most species, this deviation from random mating is due to emergence of subpopulations. Members of these subpopulations mate within their subpopulation, leading to different genetic properties. In light of recent studies on the potential impacts of ignoring these differences, we examine how to account for population substructure in both Bayesian Networks and relatedness estimation. Bayesian Networks are gaining popularity as a graphical tool to communicate complex probabilistic reasoning required in the evaluation of DNA evidence. This study extends the current use of Bayesian Networks by incorporating the potential effects of population substructure on paternity calculations. Features of HUGIN (a software package used to create Bayesian Networks) are demonstrated that have not, as yet, been explored. We explore three paternity examples; a simple case with two alleles, a simple case with multiple alleles, and a missing father case. Population substructure also has an impact on pairwise relatedness estimation. The amount of relatedness between two individuals has been widely studied across many scientific disciplines. There are several cases where accurate estimates of relatedness are of forensic importance. Many estimators have been proposed over the years, however few appropriately account for population substructure. Thus, a new maximum likelihood estimator of pairwise relatedness is presented. In addition, a novel method for relationship classification is derived. Simulation studies compare these estimators to those that do not account for population substructure. The final chapter provides real data examples demonstrating the advantages of these new methodologies.