Unraveling the low-energy conformers of neutral and charged mono- and di-saccharides with first-principles accuracy assisted by neural network potentials

We present an efficient structure sampling algorithm that combines neural network potentials (NNPs) and a well-sampled local minima dataset to explore the conformational space of mono-saccharides and di-saccharides. The structure sampling methodology leverages a "pattern transfer" approach, in which molecular initial guesses are created by utilizing existing local minima from a structurally relevant molecular system as a template. The NNP models are integrated into the structure sampling scheme to efficiently identify low-energy conformer candidates. Vibrational spectra simulated from these identified structures show qualitative alignment with experimental infrared spectra. This study demonstrates the potential of combining NNP models in an efficient structure sampling scheme to investigate flexible molecular systems, advancing understanding of carbohydrate structure-property relationships.