Menu:

Version française »

Ph.D. thesis

Bayesian methods for gene expression factor analysis

Advisor

Jean-Yves Tourneret

Co-advisor

Nicolas Dobigeon

Abstract

In the past few years, genomics has received growing interest, particularly since the map of the human genome was completed in early 2000's. Currently, medical teams are facing a new challenge: processing the signals issued by DNA chips. These signals, often of voluminous size, allow one to discover the level of a gene expression in a given tissue at any time. For example, to detect or prevent a disease in a group of observed patients based on gene expression levels innovative methods are needed to analyze the information provided by these chips.
One important problem is the identification of the specific temporal gene expression profiles of host response to a pathogen. In this problem, sensitive methods are needed that can reveal subtle and elusive expression patterns. One particularly promising approach are dimensionality reduction methods that decompose these signals into elementary patterns according to a linear mixing model. For example, PCA, ICA and manifold learning are such approaches. We propose to develop Bayesian decomposition algorithms to identify relevant biomarkers and estimate their levels of expression. The work will be based on the experience of the SC team on Bayesian statistical modeling, as well as the use of stochastic simulation methods, which have already been demonstrated for other applications (including hyperspectral imaging). Moreover, when analyzing DNA data, a crucial issue is the reduction of dimensionality of the signals to be processed. Consequently, a particular interest will be devoted to developing appropriate methods for sparse analysis and variable selection.

Collaborations

A collaboration with the team of Prof. Alfred O. Hero (Department of Electrical Engineering and Computer Science, University of Michigan, USA) is envisaged for this study. The quality of the research conducted by Prof. Hero in biostatistics will be a significant support. The methods developed will be evaluated on real signals, collected during a recent viral challenge study (2008) conducted on volunteers.