Subgraph-Based Molecular Graph Generation

De novo molecular design strategies are essential in drug discovery. With the recent advances of deep generative models and reinforcement learning in many application areas, various approaches for molecular graph generation have been proposed. In this chapter, we give the fundamental ideas of machine learning techniques for generating novel molecules and optimizing the objective for target molecules. Moreover, we introduce a new approach that focuses on subgraph structures obtained through graph mining and reassembles subgraphs guided via reinforcement learning. The subgraph-based approach can generate molecules with desired properties by presenting the trajectories of valid intermediate molecules.