Elion: A Deep Learning Based Platform for Multi-Objective Drug Hit Optimization

Here, we present Elion, a complete, deep learning-based platform for multi-objective drug hit optimization, which incorporates the generation of analog molecules, docking, and deep-learning-based binding energy estimation. The ReLeaSE molecular generator generates new molecules based on a pre-defined scaffold or similarity to a desired reference. It is also subjected to diverse constraints such as drug-likeness, synthetic accessibility, and estimated docking scores. The generated molecules can be filtered by ADMET properties, and selected molecules docked to a protein target, and finally rescored with the Deep-Learning-based DeepAtom method. Elion is 2 designed to be easy to use, and extensions with new properties or methods should be straightforward. As an example, we apply this method to generate new selective inhibitors for COX-2 (COXIBs). The source code for Elion is freely available on GitHub: https://github.com/gmseabra/elion.