Quantum Graph Neural Networks Based Protein-Ligand Classification

Graph Neural Networks (GNNs) are an emerging research area with an ongoing exploration of their practical applications. There have been promising results in combining GNNs with Quantum circuits (Quantum GNNs) for tasks such as predicting chemical and material properties, drug discovery, and analyzing complex systems such as social networks and traffic patterns. Additionally, QGNNs have the potential to provide significant speedup in certain machine-learning tasks by leveraging the power of quantum parallelism. In this research work, we propose a hybrid non-sequential Quantum Graph Neural Network (QGNN) model called QMolNet that utilizes variational quantum circuits and encoding techniques. The model is implemented for the problem of protein-ligand classification with consideration of non-bond interactions over a range of cut-off distances using the BACE dataset. Our comparative analysis shows that the proposed hybrid model outperforms various classical Graph Neural Network (GNN) models and provides state-of-the-art performance for the protein-ligand classification task. This research presents a potential direction for developing Quantum Graphical Models in quantum chemistry that incorporate chemical intuition and are comparable to existing chemical concepts and tools.