Generative Deep\nLearning-Based Efficient Design of\nOrganic Molecules with Tailored Properties

Innovative approaches to design molecules with tailored\nproperties\nare required in various research areas. Deep learning methods can\naccelerate the discovery of new materials by leveraging molecular\nstructure–property relationships. In this study, we successfully\ndeveloped a generative deep learning (Gen-DL) model that was trained\non a large experimental database (DB<sub>exp</sub>) including 71,424\nmolecule/solvent pairs and was able to design molecules with target\nproperties in various solvents. The Gen-DL model can generate molecules\nwith specified optical properties, such as electronic absorption/emission\npeak position and bandwidth, extinction coefficient, photoluminescence\n(PL) quantum yield, and PL lifetime. The Gen-DL model was shown to\nleverage the essential design principles of conjugation effects, Stokes\nshifts, and solvent effects when it generated molecules with target\noptical properties. Additionally, the Gen-DL model was demonstrated\nto generate practically useful molecules developed for real-world\napplications. Accordingly, the Gen-DL model can be a promising tool\nfor the discovery and design of novel molecules with tailored properties\nin various research areas, such as organic photovoltaics (OPVs), organic\nlight-emitting diodes (OLEDs), organic photodiodes (OPDs), bioimaging\ndyes, and so on.