Deep learning methods have gained attention in predictive process mining tasks. The most popular architectures include Recurrent Neural Networks and Convolutional Neural Networks. Inspired by the fact that Graph Neural Networks can operate directly on graph data, this work proposes a novel predictive business process monitoring method using Graph Neural Networks to learn the underlying graph representation of business processes and make predictions about the next activity of running cases. The embeddings for the events and their attributes can be learned in an end-to-end manner. Moreover, the proposed method employs a similarity function to guide the learning process and reduce overfitting. The model was evaluated using the BPIC’12 dataset and outperformed state-of-the-art methods by a large margin.