Ranking Loci for Genetic Stock Identification by Curvature Methods

Measures of the utility of loci in genetic stock identification problems are usually not based on the method of maximum likelihood, which is the actual statistical procedure used to estimate stock contributions. We present a general procedure, derived from the likelihood method, for assessing the utility of baseline data. The method depends on the curvatures of potential likelihood surfaces and can be used prior to mixture sampling. We also develop a real time implementation of a curvature measure and apply it to simulated mixture samples. The error in likelihood estimation depends on the amount of variation in genotype frequencies between reference samples as well as the location of the center of that variation. The curvature measure accounts appropriately for both factors and, in addition, is able to quantify the synergistic interaction of multiple loci. The curvature approach and simulation results are also applied to the problem of sampling allocation.