De novo drug design framework based on mathematical programming method and deep learning model

Abstract Small‐molecule drugs are of significant importance to human health. The use of efficient model‐based de novo drug design method is an option worth considering for expediting the discovery of drugs with satisfactory properties. In this article, a deep learning model is first developed for identifications of protein‐ligand complexes with high binding affinity, where the Mol2vec descriptor, the convolutional neural network, and the gate augmentation‐based Attention mechanism are used for the model construction. Then, an optimization‐based de novo drug design framework is established by integrating the deep learning model into a Mixed‐Integer NonLinear Programming (MINLP) model for drug candidate design. The optimal solution of the MINLP model is further verified by the physics‐based methods of molecular docking and molecular dynamics simulation. Finally, two case studies involving the design of anticoagulant and antitumor drug candidates are presented to highlight the wide applicability and effectiveness of the MINLP‐based de novo drug design framework.