Wavelet-Based Functional Mixed Models for Transcriptome Analysis

• Activity Number: 277
• Type: Contributed
• Date/Time: Tuesday, August 4, 2009 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Lieven Clement*+ and Ciprian M. Crainiceanu and Kristof De Beuf and Olivier Thas and Rafael Irizarry
• Companies: Ghent University and Johns Hopkins University and Ghent University and Ghent University and Johns Hopkins University
• Keywords: tiling microarray ; wavelets ; adaptive regularization ; transcript discovery ; differential expression ; ridge regression

Abstract

To get more insight in the biology of an organism, it is necessary to identify the regions of the genome that are actively transcribed. Tiling arrays can be used for this purpose. Such data are very spiky. Therefore, the wavelet based functional mixed model of Morris et al. (2006) JRSS B 68:179-199 seems a very natural framework. Their implementation is based on Bayesian MCMC methods. Due to the large dimensionality of the data, this is computationally too demanding. We introduce an alternative method, which is a leap forward in terms of numerical simplicity. Regularization is accommodated by the use of ridge regression. We exploit the link between ridge regression and mixed models to provide fast algorithms for parameter estimation and inference. In contrast with existing methods, our model can assess transcript discovery and differential expression simultaneously.