A machine learning assisted designing and chemical space generation of benzophenone based organic semiconductors with low lying LUMO energies

Abstract

Current study presents a machine learning (ML) approach to design benzophenone-based organic chromophore with their lowest possible LUMO energy (ELUMO). A dataset of their 1142 donors is collected from literature and their molecular descriptors are designed by using RDKit. Among various models, the Random Forest regression model produces accurate results to predict their ELUMO values. Based on these predictions, their 5000 new donors are designed with their Synthetic Accessibility Likelihood Index (SALI) scores. Their SHAP value analysis reveals that their electro topological state indices are the most critical descriptors to lowering ELUMOs. The top- performing donor are further extended with acceptors and their photovoltaic (PV) properties by density functional theory (DFT). Their results show their maximum open-circuit voltage (Voc) of 2.30 V, a short-circuit current (Jsc) of 47.19 mA/cm2, and a light-harvesting efficiency (LHE) of 93 %. This study demonstrates the potential of ML assisted design to design new organic chromophores.