Bayesian methods for wavelet analysis of spectral data towards biomarker discovery

Dr Chris Holmes

University of Oxford, Oxford, UK

We have developed multivariate Bayesian methods for spectral de-noising and variance decomposition into "interesting" components of variation, such as genetic, environmental and individual. The methods make use of joint inference across the sample set; rather than processing each spectra independently. The methods were developed as part of the MolPAGE consortia, where we have metabonomic data on 154 individuals (77 twin pairs) taken on multiple visits.