Hybrid quantum generative model for molecule generation

In drug discovery, generating new molecules with specific properties is time-consuming, expensive, and poses significant challenges in pharmaceutical research and development. Machine learning and deep learning technologies have successfully demonstrated their ability to accelerate drug design approaches. Additionally, quantum computing has shown promising potential for various applications, including machine learning. Currently, the combination of quantum computing and machine learning has garnered considerable attention for drug discovery problems. Recent research has developed various quantum GANs (QGANs) to generate small new molecules. Although QGANs have shown promising results in molecular generation, significant challenges remain.

In this talk, we will first introduce a novel quantum machine learning model that we developed to generate new molecules with specific physicochemical properties. We will then demonstrate how the CUDA-Q platform is employed to accelerate the training of our model by parallelizing the training data across multiple nodes and GPUs, allowing us to use large batch sizes and a substantial amount of data points for training. Finally, we will provide detailed analyses comparing the classical and quantum models. Our model, utilizing a large number of data points, enables a more effective comparison against the classical machine learning model.