Prediction of organic molecular optical absorption energy based on deep learning using mordred descriptor features

Abstract

Organic molecules have several characteristics based on optical absorption. A molecule absorbs a specific light wavelength that represents electronic energy in the material. The absorption wavelength is also related to atomic binding of molecules that interact with material during light exposure producing electronic instability in the material. Molecular absorption can be measured using several optical spectrometer configurations that contain a light source, optical path, and light detector. The measurements of molecular optical absorption have been reported in several research and have been collected into a material database. In this paper, we developed a model for predicting organic molecular optical absorption using deep learning. The model can predict molecular absorption energy based on extracted features of molecular structure using Mordred descriptor features extraction. Our model used 1625 molecules of absorption in several solvent datasets that split into 80% training and 20% testing dataset. The result show that our model has a good agreement with experimental data with correlation coefficient 0.96 and mean average error 0.172 eV.