Evaluation of statistical methods for the analysis of forensic DNA mixtures

Analysis of forensic DNA mixtures recovered from crime scenes is one of the most challengingtasks in forensic science. DNA mixture raise two main questions: “how many contributors arethere” and “what are the genotypes of the contributing individuals?” The genetic characterizationalone of such samples does not always answer these questions. In fact, whenever more thantwo alleles are observed at a given locus, several distinct genotypic combinations are plausiblefor the unknown contributors to the sample, and it is not possible to determine the number ofthese contributors with absolute certainty. Besides, the presence of anomalies related to DNAtyping techniques, such as contamination or allele loss (drop-out), can further complicate theanalysis.Numerous statistical developments facilitating DNA mixtures interpretation were proposed,but they did not receive the expected success in the forensic community. The main explanationfor this is that these methods are not validated for forensic casework.In order to achieve this validation criterion, the methods must undergo a rigorous evaluationstep. The latter raises two questions: i) how methods should be evaluated? and ii) whattools can be used to conduct evaluation studies? In this thesis we attempt to answer bothquestions. First, we evaluate methods dedicated to two key issues, the estimation of the numberof contributors to DNA mixtures and the estimation of drop-out probabilities. Second, wepropose an “open-source” software that offers a number of functionalities dedicated to facilitatingmethod evaluation through the simulation of data commonly encountered in forensic settings.This thesis aims to provide a concrete answer to the issues raised by forensic DNA mixtures,by providing a methodology for method evaluation and by offering necessary tools to enablemethod evaluation.