Identification of Bioisosteric Substituents by a Deep Neural Network

Bioisosteric design is a classical technique used in medicinal chemistry to improve potency, druglike properties, or the synthetic accessibility of a compound or to find similar potent compounds that exist in novel chemical space. Bioisosteric design involves replacing part of a molecule by another part that has similar properties. Such replacements may be identified by applying medicinal chemistry knowledge, by mining chemical databases or by choosing analogues similar in molecular physicochemical properties. In this article, a novel approach to identify bioisosteric analogues is described where the suggestions are made by a deep neural network trained on data collected from a large corpus of medicinal chemistry literature. The network trained in this way is able to mimic the decision making of experienced medicinal chemists and identify standard as well as nonclassical bioisosteric analogues, even for the structures outside the training set. Examples of the results are provided and application possibilities are discussed.