Pharmacogenomics (PGx) studies the influence of the genome in drug response, with knowledge units of the form of ternary relationships genomic variation-drug-phenotype. State-of-the-art PGx knowledge is available in the biomedical literature as well as in specialized knowledge bases. Additionally, Electronic Health Records of hospitals can be mined to discover such knowledge units that can then be compared with the state of the art, in order to confirm or temper relationships lacking validation or clinical counterpart. However, both discovering and comparing PGx relationships face multiple challenges: heterogeneous descriptions of knowledge units (languages, vocabularies and granularities), missing values and importance of the time dimension. In this research, we aim at proposing a framework based on Semantic Web technologies and Formal Concept Analysis to discover, represent and compare PGx knowledge units. We present the first results, consisting of creating an integrated knowledge base of PGx knowledge units from various sources and defining comparison methods, as well as the remaining issues to tackle.