Decoding the Signatures of Chromophore-Solvent Interactions in InfraRed Spectroscopy with Machine-Learning: Insights from a Hybrid Density-Functional Theory/Molecular Mechanics Dynamics Model

We report a novel, chemically intuitive, machine-learning-based approach for assigning vibrational bands in complex molecular systems studied by infrared (IR) spectroscopy. As a complementary alternative to traditional power spectrum analysis, this method accelerates vibrational mode assignment by decomposing the IR spectrum into contributions from molecular fragments rather than analyzing atom-by-atom contributions. We demonstrate the effectiveness of this approach through a case study involving a chromophore in noncovalent interaction with a solvent molecule. Specifically, we show that it rapidly reveals the IR signature associated with the hydrogen-bond interaction.