Estimation of mutation rates from paternity cases using a Bayesian network

We present a statistical model and methodology for making inferencesabout mutation rates from paternity casework. This takes properaccount of a number of sources of potential bias, including hiddenmutation, incomplete family triplets, uncertain paternity status anddiffering maternal and paternal mutation rates, while allowing a widevariety of mutation models. A Bayesian network is constructed tofacilitate computation of the likelihood function for the mutationparameters. It can process both full and summary genotypicinformation, from both complete putative father-mother-child tripletsand defective cases where only the child and one of its parents areobserved.Detailed analysis of a specific dataset is used to illustrate theeffects of the various types of biases, and of the assumed mutationmodel, on inferences about mutation parameters.