Explainable AI (XAI) is a research area whose objective is to increase\ntrustworthiness and to enlighten the hidden mechanism of opaque machine\nlearning techniques. This becomes increasingly important in case such models\nare applied to the chemistry domain, for its potential impact on humans'\nhealth, e.g, toxicity analysis in pharmacology. In this paper, we present a\nnovel approach to tackle explainability of deep graph networks in the context\nof molecule property prediction t asks, named MEG (Molecular Explanation\nGenerator). We generate informative counterfactual explanations for a specific\nprediction under the form of (valid) compounds with high structural similarity\nand different predicted properties. Given a trained DGN, we train a\nreinforcement learning based generator to output counterfactual explanations.\nAt each step, MEG feeds the current candidate counterfactual into the DGN,\ncollects the prediction and uses it to reward the RL agent to guide the\nexploration. Furthermore, we restrict the action space of the agent in order to\nonly keep actions that maintain the molecule in a valid state. We discuss the\nresults showing how the model can convey non-ML experts with key insights into\nthe learning model focus in the neighbourhood of a molecule.\n