Domain experts in various fields have formalised their knowledge in ontologies. These are especially present in biomedical sciences where such controlled vocabularies are also used to annotate the data in various databases. This means that knowledge graphs with information from various sources can be formed, accompanied by ontologies describing the classes used and grounding the problem in knowledge acquired over years of research. This project will investigate neuro-symbolic methods to learn and predict from knowledge graphs, e.g., through learning embeddings adhering to the rules stated in the ontologies. By utilising such rules search spaces can be reduced, asserting predictions are viable, as well as making the models and predictions more interpretable by grounding them in semantic meaningful representations. Initially the goal with this project, apart from developing neuro-symbolic methods for knowledge graph learning, is to generate hypotheses for automated Robot Scientists performing experiments on yeast. This will be done by predicting missing information in knowledge graphs formed from available information related to the organism.