A machine learning vibrational spectroscopy protocol for spectrum prediction and spectrum-based structure recognition
Hao Ren, Hao Li, Qian Zhang, Lijun Liang, Wenyue Guo, Fang Huang, Yi Luo, Jun Jiang
Abstract
Vibrational spectroscopy is one of the most commonly applied techniques for determining molecular structures. Conventional applications often involve extensive expertise or expensive first-principles computational effort in order to establish one-to-one spectrum-structure relationships. Here we developed a machine-learning protocol to correlate spectral fingerprints with local molecular structures. Our protocol enables not only quick and accurate prediction of infrared (IR) absorption and Raman vibrational spectra based on molecular structures, but more importantly, also enables structure recognition of chemical groups from vibrational spectral features. IR and Raman spectral features arising from different selection rules were recurrently fed to the model to achieve a nearly zero error rate in structure recognition. Both the spectrum prediction and structure recognition models have good transferability, implying a high possibility of being extended to various spectral or non-spectral characteristics. This machine learning protocol may provide impovements to real-time field applications in many areas of spectroscopy.