Prediction of the\nInfrared Absorbance Intensities\nand Frequencies of Hydrocarbons: A Message Passing Neural Network\nApproach

Accurately\nand efficiently predicting the infrared (IR)\nspectra\nof a molecule can provide insights into the structure–properties\nrelationships of molecular species, which has led to a proliferation\nof machine learning tools designed for this purpose. However, earlier\nstudies have focused primarily on obtaining normalized IR spectra,\nwhich limits their potential for a comprehensive analysis of molecular\nbehavior in the IR range. For instance, to fully understand and predict\nthe optical properties, such as the transparency characteristics,\nit is necessary to predict the molar absorptivity IR spectra instead.\nHere, we propose a graph-based communicative message passing neural\nnetwork algorithm that can predict both the peak positions and absolute\nintensities corresponding to density functional theory calculated\nmolar absorptivities in the IR domain. By modifying existing spectral\nloss functions, we show that our method is able to predict with DFT-accuracy\nlevel the IR molar absorptivities of a series of hydrocarbons containing\nup to ten carbon atoms and apply the model to a set of larger molecules.\nWe also compare the predicted spectra with those generated by the\ndirect message passing neural network. The results suggest that both\nalgorithms demonstrate similar predictive capabilities for hydrocarbons,\nindicating that either model could be effectively used in future research\non spectral prediction for such systems.