This thesis studies the use of ontology and knowledge base for guiding various steps of the Knowledge Discovery in Databases (KDD) process in the domain of pharmacogenomics. Data related to this domain are heterogeneous, complex, and disseminated through several data sources. Consequently, the preliminary step that consists in the preparation and the integration of data is crucial. For guiding this step, an original approach is proposed, based on a knowledge representation of the domain within two ontologies in description logics : SNP-Ontology and SO-Pharm. This approach has been implemented using semantic Web technologies and leads finally to populating a pharmacogenomic knowledge base. As a result, data to analyze are represented in the knowledge base, which is a benefit for guiding following steps of the knowledge discovery process. Firstly, I study this benefit for feature selection by illustrating how the knowledge base can be used for this purpose. Secondly, I describe and apply to pharmacogenomics a new method named Role Assertion Analysis (or RAA) that enables knowledge discovery directly from knowledge bases. This method uses data mining algorithms over assertions of our pharmacogenomic knowledge base and results in the discovery of new and relevant knowledge.