The exploration of drug repurposing as an innovative and efficient strategy for discovering new treatments across a spectrum of diseases has gained considerable attention in the last few years. One of the major fields in in-silico drug repurposing utilizes Knowledge Graphs (KGs) as powerful tools in identifying potential drug candidates. In this work, we make use of the NeDRex KG [1] to build a pipeline/workflow that can be used by researchers to generate drug candidates. These candidates are produced by training a Graph Neural Network (GNN) known as GraphSAGE [2], that achieves high performance metrics (AUC > 0.9 and AUPR > 0.9) in the validation and training set. To validate these predictions, several diseases from different categories (metabolic, infectious, rare diseases, viral...) were selected and a literature search was performed to see if the predicted drug-disease association had been previously reported. One of the major novelties of this work is a new XAI technique that creates hypotheses that may help to support the predictions. This XAI method works by analyzing all simple paths between the drug and the disease; each simple path is then scored by passing it through the GNN, being the path with the highest score the one that contains the most significant nodes and edges to the GNN. Scoring each simple path may require a significant amount of time; however, it is possible to accelerate the method by aggregating similar paths together. This is done using the MinHash algorithm. Finally, to validate the generated hypotheses we make use of the DrugMech database [3], which contains several curated drug-disease pairs with their corresponding explanation. The validation is done by comparing the hypotheses obtained with the ones appearing in DrugMech. The main limitation, however, is that not all nodes/paths present in DrugMech are present in the NeDRex graph. The results look promising (Hit@1 score: 0.3, Hit@3 score: 0.4, Hit@5 score: 0.5) but we are now exploring other ways to evaluate the XAI method.