<p>Nowadays, news can be intentionally created or manipulated to amplify social influence. As a consequence, the proliferation of fake news has become a significant challenge leading to substantial negative impacts. This phenomenon highlights the urgent need for robust methods for fake news detection and prevention. Although several approaches have been proposed, only a few have focused on capturing the propagation patterns of fake news. Therefore, this paper proposes a novel approach based on Metric Learning and Graph Neural Networks for fake news detection. First, in this approach, we create a set of propagation graphs from news articles and then utilize Bidirectional Encoder Representations from Transformers (BERT) to generate node embeddings to capture semantic understanding of news topics. Second, we employ a graph neural network architecture to capture the structural information within the propagation graphs and generate graph embeddings. Third, we leverage a metric learning method based on the triplet loss to enforce a clear distinction among the graph embeddings belonging to different classes. Finally, we feed these improved embeddings into a fake news classifier. The entire architecture of the proposed approach is trained in an end-to-end manner. To demonstrate the effectiveness of our proposal, we conducted extensive experiments on four real-world benchmark datasets covering multiple languages and topics. The results show that our approach significantly outperforms state-of-the-art methods.</p>