We worked on clique minimal separator decomposition. In order to compute this decomposition on a graph G we need to compute the minimal separators of its triangulation H. In this context, the first efforts were on finding a clique minimal separators in a chordal graph. We defined a structure called atom tree inspired from the clique tree to compute and represent the final products of the decomposition, called atoms. The purpose of this thesis was to apply this technique on biological data. While we were manipulating this data using Galois lattices, we noticed that the clique minimal separator decomposition allows a divide and conquer approach on Galois lattices. One biological application of this thesis was the detection of fused genes which are important evolutionary events. Using algorithms we produced in the course of along our work we implemented a program called MosaicFinder that allows an efficient detection of this fusion event and their pooling. Another biological application was the extraction of genes of interest using expression level data. The atom tree structure allowed us to have a good visualization of the data and to be able to compute large datasets.
