Contribution of biostatistical methods to genomic and proteomic data analysis : a case for microarray data analysis, transcriptional network inference, and protein binding site detection

(eng) The main scope of this thesis is the application of biostatistical methods to genomic and proteomic data analyses. Each part deals specifically with one bioinformatics subfield. In the first part, we examine the preprocessing of spotted gene expression microarray data. Gene expression microarrays can measure simultaneously the expression levels of thousands of genes. As the microarray signals are usually very noisy, there is a clear need to develop statistical methods and bioinformatics tools discriminating more efficiently a specific signal from the background noise. Accordingly, the current work has contributed to develop tools that include a new method combining microarray data acquired through multiple laser scans, a new hybrid transformation as well as an outlier detection method for microarray data normalization. The second part of this thesis, also related to system biology discipline, focuses on the development of computational methods for reconstructing transcriptional networks from microarray data and functional similarities. The third part of this thesis focuses on a structural analysis of protein-protein interactions. In this last part, signal processing and statistical tools are used to develop a method predicting the locations of sites of interest on the protein surfaces. The method is used for mapping the location of binding sites on various protein surfaces. Each part of the thesis pinpoints therefore the contributions of these original biostatistical methods and bioinformatics tools by carrying out an extensive comparison of their respective performances with those of existing methods.