| This dissertation sets forth a foundation for a continuous model for the interpretation of DNA mixture evidence. We take a new approach to modelling electropherogram data by modelling the actual electropherogram as a curve rather than modelling the allelic peak areas under the curve. This shift allows us to retain all the data available and to bypass the approximation of peak areas by GeneMapperRTM (Applied Biosystems, 2003). The two problems associated with the use of this programme---prohibitive costs and patented processes---are thus avoided.;With a model in place for single-source DNA samples, we develop an algorithm that deconvolves a two-person mixture into its separate components and provides the posterior probabilities for the resulting genotype combinations.;In addition, because of the widely recognized need to perform further research on continuous models in mixture interpretation and the difficulty in obtaining the necessary data to do so (due to privacy laws and laboratory restrictions), a tool for simulating realistic data is of the utmost importance. PCRSIM (Gill et al., 2005) is the most popular simulation software for this purpose. We propose a method for refining the parameter estimates used in PCRSIM in order to simulate more accurate data.;To establish a model for electropherogram data, we explore two Bayesian wavelet approaches to modelling functions (Chipman et al., 1997; M. Clyde et al., 1998) as well as a Bayesian Adaptive Regression Splines approach (DiMatteo et al., 2001). Furthermore, we establish our own genotyping algorithm, once again circumventing the need for GeneMapperRTM, and obtain posterior probabilities for the resulting genotypes. |
