Deep learning-based inverse design model for intelligent discovery of organic molecules

Abstract: The discovery of high-performance functional materials is crucial to overcome technical issues of modern industries. And much effort has been exerted to accelerate and facilitate the process not only in experiment but also in materials design. Recently, machine learning has been paid much attention since it can provide rational guides for efficient materials exploration without resorting to time-consuming iterations and prior human knowledge. In this regard, we developed an inverse design model based on deep encoder-decoder architecture for targeted molecular design. Inspired by neural machine language translation, the deep neural network encoder extracts hidden features between molecular structures and their materials properties while the recurrent neural network decoder reconstructs the extracted features into new molecular structures having target properties. In materials design tasks, the fully data-driven methodology successfully learned design rules from given databases, and generated promising light-absorbing molecules and host materials for phosphorescent organic light-emitting diode by creating novel ligands and combinatorial rules.