Methods to classify familial relationships in the presence of laboratory errors, without parental data

We consider the problem of accurate classification of family relationship in the presence of laboratory error without parental data. We first propose an adjusted version of the test statistic proposed by Ehm and Wagner based on the summation over a large number of genetics markers. We then propose use of the Bayes factor as a classification rule. We prove theoretically that the Bayes factor is the optimal classification rule in that the total classification error is minimized. We show via simulations that both the adjusted Ehm and Wagner method and Bayes factor classification rule reduce misclassification errors, and that the Bayes factor classification rule is robust against under-estimation or over-estimation of laboratory errors. For monozygotic twins versus dizygotic twins, the correct classification rate of the Bayes rule is over 99%. For full-siblings versus half-siblings, the Bayes factor classification rule generally outperforms Ehm and Wagner's method (in Am J Hum Genet 62:181-188, 1998), especially when full-sibling proportion is high.