In this thesis, we show the importance of the modeling and the cooperation of metaheuristics for solving real problems in Bioinformatics. Two problems are studied : the first in the Proteomics domain for the protein identification from spectral data analysis and the second in the domain of the structural analysis of molecules for the flexible molecular docking problem. So, for the first problem, a new model has been designed based on a direct comparison of a raw experimental spectrum with protein from databases. This model has been included in an identification engine by peptide mass fingerprinting called ASCQ_ME. From this model, an approach for the " de novo protein sequencing " problem has been proposed and validated. In this problem, a protein sequence has to be found with only spectral information. Our model is a three step approach called SSO for " Sequence ", " Shape " and " Order ". After a study of each step, SSO has been implemented and tested with three metaheuristics collaborating sequentially. For the second problem, a study of new multi-objective models has been made and has allowed to design eight different models tested with parallel multi-objective genetic algorithms. Twelve configurations of genetic operators has been tested in order to prove the efficiency of the hybridizing of genetic algorithms with local searches. For each part of this work, the ParadisEO platform has been used and more particularly the ParadisEO-MO part dedicated to single solution based metaheuristics for which we have substantially contributed. All this work has been funded by the " PPF BioInformatique " of the " Université des Sciences et Technologies de Lille " and by the ANR Dock project.
