This talk will introduce Graph Neural Networks (GNNs) as a powerful toolfor the design and evaluation of new targeted drugs.Drug Discovery is a fundamental but expensive process to make newpharmacological products available for healthcare. The generation ofmolecular graphs is of particular interest for drug discovery, as it couldprovide a technique for designing large amounts of possible drugcandidates. GNNs can be used as molecular graph generators, able to createnew drug-like molecules from scratch, which can also be adapted to fit inspecific pockets of the protein surface. Indeed, constrained generationallows for designing molecules with both the desired chemical andstereochemical properties, while their side-effects can be evaluated basedon their molecular structure. Moreover, the introduction of Composite GNN(CGNN) models, designed for processing heterogeneous graphs, has allowedthe study of even more complex networks. With CGNNs, drug side-effects canbe predicted based on a graph describing the interactions between drugsand human genes.Actually, a three-step chain can be devised, in which the graph generatorconstitutes the first step, aimed at producing a large pool of possibledrug candidates. The drug candidates could then be screened for theirdrug-likeness, retaining only compounds with a high druggability score.Finally, the selected compounds can be evaluated, filtering out those withtoo many or too dangerous side-effects. In this way, discovering new drugscan be carried out as a total in-silico procedure, before clinical trials,saving costs in terms of time and money.