The 85th Annual Meeting of the American Association of Physical Anthropologists (2016)

rASUDAS: A New Method for Estimating Ancestry from Tooth Crown and Root Morphology


1Department of Anthropology, University of Nevada Reno, 2Laboratory of Forensic Anthropology, University of Coimbra, 3Research Centre in Evolutionary Anthropology and Paleocology, Liverpool John Moores University

April 14, 2016 , Atrium Ballroom A/B Add to calendar

Based on crown and root trait frequencies reported in The Anthropology of Modern Human Teeth, an application was developed that assigns individuals to a geographic subdivision of humankind. All that is required is to take an individual dentition and score traits as present (at or above a given breakpoint), absent, or unobservable. The analysis generates the probability of an individual being assigned to one or more groups. The method was developed in two stages. First, Nei’s distance matrix was computed using each crown and root trait, from which a hierarchical clustering tree was created using UPGMA algorithm with complete linkage. Based on a visual inspection of the clustering tree, two to seven biogeographic population clusters were defined. Second, probabilistic biogeographic ancestry prediction models were fitted using naive Bayes classifier algorithm, a simple yet powerful technique that uses Bayes’ theorem as a prediction engine. This algorithm outputs the ancestral group and its associated posterior probability. It is called naive because the algorithm assumes total conditional independence between traits, which significantly simplifies the full multivariate predictive density computation. Mathematical conditional independence is a strong assumption, but this conforms to the working assumption that crown and root traits are expressed independently of one another. To simplify and expand the usage of this method, a simple program and web application named rASUDAS was developed. Test runs on 150 data sheets from world populations arrived at correct classifications ranging from 57 to 92 percent, depending on the number of biogeographic groups included in the analysis.