Machine Learning-Based Multispectral Fusion for Analyzing Molecular Structural Features

Spectroscopy is a fundamental tool for analyzing the molecular structure of substances. Traditional spectral interpretation methods heavily rely on the expertise of spectroscopic analysts, which can be labor-intensive and susceptible to errors, particularly when analyzing complex molecules that lack prior documentation in the literature. In this study, we introduce an innovative approach that autonomously identifies the presence and quantifies the proportions of various substructures within a molecule using its infrared (IR), Raman, and nuclear magnetic resonance (NMR) spectra. This method eliminates the need for database searches or expert-defined rules, offering a more efficient and objective alternative. Experimental results demonstrate that the integration of multiple spectral techniques provides a more comprehensive and accurate representation of a molecule's structural information. Additionally, the transfer performance of our models on external data sets underscores the robustness and generalizability of multispectral integration for molecular structure inversion.