Spaces:
Running
Update README.md
Motivation
It is a time-consuming and costly process to develop affordable and high-performance organic photovoltaic materials. Developing reliable computational methods to predict the power conversion efficiency (PCE) is crucial to triage unpromising molecules in large-scale databases and accelerate the material discovery process. In this study, a deep learning-based framework (DeepAcceptor) has been built to design and discover high-efficient small molecule acceptor materials. Specifically, an experimental dataset was constructed by collecting data from publications. Then, a BERT-based model was customized to predict PCEs by taking fully advantages of the atom, bond, connection information in molecular structures of acceptors, and this customized architecture is termed as abcBERT. The computation molecules and experimental molecules were used to pre-train and fine-tune the model, respectively. The molecular graph was used as the input and the computation molecules and experimental molecules were used to pretrain and finetune the model, respectively. In sum, DeepAcceptor is a promising method to predict the PCE and speed up the discovery of high-performance acceptor materials.