Bayesian Methods for Copy Number Inference in Heterogeneous Cancers Using SNP Arrays

• Activity Number: 360
• Type: Contributed
• Date/Time: Tuesday, August 3, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Christopher Yau*+ and Christopher C. Holmes
• Companies: University of Oxford and University of Oxford
• Keywords: Bayesian ; Cancer ; Copy Number ; Nonparametric ; Dirichlet Process ; Decision Theory

Abstract

We present a nonparametric Bayesian Hidden Markov approach to the problem of inferring copy number alterations within heterogeneous cancer samples designed to capture the full complexity of the structural variations observed in cancer as well as genetic heterogeneity within the tumour and unknown ploidy. Our noise model is complexity of the noise process we include a nonparametric mixture of Dirichlet Process (MDP) model.

We also incorporate formal methods for decision making into our models under a Bayesian decision theoretic approach using a class of loss functions which incorporate errors on the breakpoints (the locations of DNA deletion-duplication events) which amounts to a loss on the state transitions of the (hidden) state sequence. We show that reporting under this particular loss (or utility) function better reflects the reality of how the models are used in practice than conv