Graph convolutional neural networks (GCNNs) have emerged as powerful tools for predicting molecular properties in chemistry. However, their black-box nature poses challenges for interpretability, hindering their widespread adoption. In this work, we propose a symmetry-sensitive method for interpreting GCNN models, aiming to provide explanations that align with chemical intuition while maintaining computational efficiency. We introduce the MolgraphX explainer method, tailored to highlight the importance of specific molecular substructures in predictions. We demonstrate the effectiveness of our approach using multiple data sets of small organic molecules with different properties. Our method offers insights into the underlying chemical mechanisms, bridging the gap between formal accuracy and chemical intuition. Through extensive experimentation, we validate the efficacy and efficiency of our proposed method, offering chemists a valuable tool for understanding and interpreting GCNN predictions for molecular chemistry applications.
Karpov et al. (Sat,) studied this question.