Inferring ethnicity using 15 autosomal STR loci--comparisons among populations of similar and distinctly different physical traits.

The forensic utility of short tandem repeat (STR) polymorphisms for ethnic inference (EI) has recently received limited attention from the research community. The large volume of population specific STR data available from public literature affords us the opportunity to assess the reliability of EIs based on STR profiles, and present significant results. We applied a binary Bayesian approach using 15 autosomal STR loci from computer-simulated and real STR profiles of Han Chinese, Japanese, Korean, American Caucasian, and a South Asian Indian aboriginal tribal group called Lodha. A correct classification rate of > or = 90% was achieved via simulation for each East Asian group (Han Chinese, Japanese, and Korean) when paired with the American Caucasian or the tribal Indian group. Similar result was obtained between the Caucasian and the Indian group, while the correct classification rates among the three East Asian groups were much lower (approximately 70%). Results based on real profiles closely matched those achieved via simulation, establishing the credibility of our classifier performance metrics. This study shows that binary Bayesian classifier performed best for ethnic groups with distinctly different physical traits. This observation might be due to accidental differences caused by bottlenecks, genetic drift, etc. However, it could also be taken as an indication that STRs, commonly referred to as "junk" DNA, may have an effect on phenotype, a possibility that warrants further research.