Generating a chemical database of organic nanomers and applying active learning to predict HOMO, LUMO and band gap: Accelerating optoelectronic nanopolymer materials discovery.

Abstract

Organic nanogrids are versatile molecular hornstones and nanoplatforms of organic high-dimensional, low-entropy materials. It is urgent to construct virtual databases of organic nanomers for accelerating the discovery and performance optimization of novel 0/1/2/3-dimensional nanopolymer optoelectronic materials. In this study, we generated a comprehensive dataset of 11,224 ladder-type gridarenes, covering a wide range of chemical compositions and structural variations. A random selection of 220 small sample sets was aggregated, and fragment-level constrained density functional theory (CDFT) was employed to extract molecular descriptors. These descriptors were then used to train machine learning models with high predictive accuracy for band gap, highest occupied molecular orbital (HOMO), and lowest unoccupied molecular orbital (LUMO) energies (the coefficient of determination values of 0.94, 0.92, and 0.87, respectively). During the active learning process, 3112 representative gridarenes were iteratively selected from our 11,224-compound library, refining band-gap predictions to a mean absolute error below 0.11 eV. This process pinpointed top candidates for blue-light emission and demonstrated an accelerated, data-driven route to next-generation organic optoelectronic nanomaterials.