In the era of artificial intelligence (AI), deep learning has led to remarkable advancements in various fields, with examples such as AlphaFold [1], GPT-4 [2] or the very recent AlphaMissense [3]. In all these advances, all input data had a tabular nature, i.e., all training examples were expressed as a regular, component-based, vector of values. However, as we confront modeling problems where some of the central data structures go beyond that static and rigid structure to become graphs, the need for specialized techniques emerges. Graph Neural Networks (GNNs) have been advancing rapidly in recent years, improving the results of classical deep learning techniques by combining the use of tabular data with graph structures, processing data that can be represented as graphs, such as social or citation networks. In the field of biology, these models are rapidly improving results in a multitude of tasks, such as the prediction of unlabeled proteins or the generation of synthetic molecules through actual molecular graph learning [4], prediction of protein-phenotype associations with protein-protein associations networks [5], or even for drug-target interactions [6].