Assesing performance of PCR analysis using Bayesian modelling

Licentiate thesis, 2014

Different analytical methods are available to analyse the DNA samples from crime scenes or to screen samples for the presence of some target bacterium. For this purpose a method known as Polymerase Chain Reaction (PCR) can often be used. In the two papers of this thesis we have investigated the performance of PCR analysis in different contexts using Bayesian modelling. In the first paper in relation to analyses of crime scene samples containing low amount of human DNA and in the second paper in relation to analyses of samples potentially containing Bacillus anthracis, the bacterium causing anthrax. Generally, such analyses are made to test some hypothesis about the true state of nature. As for analyses of crime scene samples with human DNA the hypothesis to be tested is often about the source of the DNA. As for the investigation of samples for dangerous bacteria the hypotheses can e.g. be about the presence, amount and distribution of the bacteria or about the source of the outbreak; important questions for investigation of e.g. bioterrorism or feed- and food safety. We propose some new models for assessing the performance of PCR analysis of microbiological samples and for the interpretation of results. These are based on Bayesian hierarchical modelling, taking the relevant dependencies in data within and between experiments into account. The posterior distributions of the model parameters are assessed using Gibbs sampling via the software OpenBUGS. The adequacies of the final models are checked by comparison of model predictions to observed data. In the first paper we propose new PCR analysis settings for low-template DNA samples. We also find differences in the drop-out probability between several markers. In the second paper we compare different approaches to assess the probability of detection for B. anthracis in new materials using B. cereus as a surrogate for B. anthracis.