Learning $M$-Order Spectrum Graphs to Identify Unknown Chemical Compounds from Infrared Spectroscopy Data

Infrared (IR) spectrum analysis is an efficient and generally applicable method for identifying unknown chemical compounds. However, the analysis quality of the IR spectrum essentially depends on a labor-intensive manual search of human experts. We propose a graph-based machine learning approach for automatically identifying unknown chemical compounds from the input IR spectra without the manual analysis of human experts. To this end, we define a graph representation of unstructured spectrum data and devise a sequence model based on graph neural networks. We evaluated the proposed method in three common tasks of the IR spectrum analysis: material class classification, functional group detection, and compound identification. In these experiments, the proposed method achieved state-of-the-art prediction accuracy. In particular, the proposed method showed a prediction accuracy of 41.82%-85.10% in identifying unknown chemical compounds from their IR spectra, which is one of the most challen]ging tasks in chemical analysis.