Learning to Extend Molecular Scaffolds with Structural Motifs

Recent advancements in deep learning based modelling of molecules promise to accelerate in silico drug discovery. There is a plethora of generative models available, which build molecules either atom-by-atom and bond-by-bond or fragment-by fragment. Apart from property-driven generation, many drug discovery projects also require a fixed scaffold to be present in the generated molecule, and incorporating that constraint has been recently explored. In this work, we present a new graph based model that learns to extend a given partial graph by flexibly choosing between adding individual atoms and entire fragments. Our model does not assume access to a predefined vocabulary of scaffolds; instead, extending a scaffold is implemented by using it as the initial partial graph. This is only possible because our model does not depend on generation history, and has been trained to generate molecules using a variety of generation orders. We show that using a randomized generation order is necessary for good performance when extending scaffolds, and that results are further improved when increasing motif vocabulary size