Sep 2, 2020 – 11:00 AM
The generation of graph–structured data is an emerging problem in the field of deep learning. Various solutions have been proposed in the last few years, yet the exploration of this branch is still in an early phase. Graph data being ubiquitous in real world applications, efficient graph generators could provide a good solution for many different problems. One very relevant application of this class of models is the discovery of new drug molecules, which are naturally represented as graphs. In this paper, we introduce a molecular graph generator based on multiple Graph Neural Network modules. Our model is trained on the QM9 molecule dataset and tested in an unconditional generation session. The results show that our model is capable of generalizing molecular patterns seen during the training phase, without overfitting. Generated molecules show very good validity, uniqueness and novelty scores, while still retracing very closely the property distributions measured on a held–out test set of QM9 molecules.