Abstract
Graph neural networks (GNNs) have emerged as powerful tools to accurately predict materials and molecular properties in computational and automated discovery pipelines. In this article, we exploit the invertible nature of these neural networks to directly generate molecular structures with desired electronic properties. Starting from a random graph or an existing molecule, we perform a gradient ascent while holding the GNN weights fixed in order to optimize its input, the molecular graph, towards the target property. Valence rules are enforced strictly through a judicious graph construction. The method relies entirely on the property predictor; no additional training is required on molecular structures. We demonstrate the application of this method by generating molecules with specific energy gaps verified with density functional theory (DFT) and with specific octanol-water partition coefficients (logP). Our approach hits target properties with rates comparable to or better than state-of-the-art generative models while consistently generating more diverse molecules. Moreover, while validating our framework we created a dataset of 1617 new molecules and their corresponding DFT-calculated properties that could serve as an out-of-distribution test set for QM9-trained models.