Approximate Bayesian Computation - with applications to breast cancer data
Specialeforsvar ved Laura Sørensen
Titel: Approximate Bayesian Computation - with applications to breast cancer data
Abstract: This thesis addresses the likelihood-free simulation method called approximate Bayesian computation (ABC). The study describes the foundation of ABC and examines two extensions of ABC. The first extension shows that under the assumption of errors, ABC performs exact simulation using the error distribution as a weighting scheme. The second extension shows that ABC is improved by adjusting the samples using local linear regression. ABC is illustrated by inferring the mutation rate and the error rates for data from a breast cancer tumour. A method for evaluating the quality of the applied summary statistics is suggested. The final assessment of the approximation in ABC shows that the regression adjustment method is less sensitive to the choices of tolerance level, number of summary statistics and prior distribution than the basic ABC method
Vejleder: Carsten Wiuff
Censor: Asger Hobolth, Aarhus Universitet