Informing biological models for probabilistic methods of DNA profile interpretation

Abstract

The interpretation of forensic DNA profiles is complicated when they are poor quality or have low quantities of DNA, and the presence of DNA from more than one individual. There is a significant diversity in the methods used by laboratories to interpret profiles. These methods use different amounts of the data available within the profile. This diversity has been shown to lead to varying or contradictory conclusions being drawn from the same evidence. Probabilistic methods are being advocated as the way forward as they use more of the information within the profile and they help improve with the consistency of results between analysts. Barriers to the uptake of probabilistic software within forensic laboratories include a lack of understanding of the methods and uncertainties in the limits of the software and software implementation requirements. This thesis uses statistical methods to develop biologically informed models to assist with DNA interpretation of both autosomal and lineage markers. Models have been developed within this thesis using empirical data to predict the expected height of allelic and stutter peaks. These models are intended for use within probabilistic software. A number of topics relating to the implementation of such software are examined. In addition, the effect of linked loci on match probabilities is explored. This work has previously been published as 16 separate papers which have been reproduced within this thesis as one cohesive body of work. A majority of the models have also now been implemented in the commercial software product “STRmix” which is currently being sold worldwide.