Deep Learning\nfor Generating Phase-Conditioned Infrared\nSpectra

Infrared (IR) spectroscopy is an efficient method for\nidentifying\nunknown chemical compounds. To accelerate IR spectrum analysis, various\ncalculation and machine learning methods for simulating IR spectra\nof molecules have been studied in chemical science. However, existing\ncalculation and machine learning methods assumed a rigid constraint\nthat all molecules are in the gas phase, i.e., they overlooked the\nphase dependency of the IR spectra. In this paper, we propose an efficient\nphase-aware machine learning method to generate phase-conditioned\nIR spectra from 2D molecular structures. To this end, we devised a\nphase-aware graph neural network and combined it with a transformer\ndecoder. To the best of our knowledge, the proposed method is the\nfirst IR spectrum generator that can generate the phase-conditioned\nIR spectra of real-world complex molecules. The proposed method outperformed\nstate-of-the-art methods in the tasks of generating IR spectra on\na benchmark dataset containing experimentally measured 11,546 IR spectra\nof 10,288 unique molecules. All implementations of the proposed method\nare publicly available at https://github.com/ngs00/PASGeN.