Nuclear spin-spin coupling constants prediction based on XGBoost and LightGBM algorithms

Nuclear magnetic resonance (NMR) is a robust method for the analysis of molecular complex structures, and the measurement of the nuclear spin-spin coupling constant is the key. In this paper, based on the 3D coordinates of the atoms in the molecule, the spin-spin coupling constants of atom-pairs are directly predicted using Extreme Gradient Boosting (XGBoost) and Light Gradient Boosting Machine (LightGBM). The calculated result of DFT method is taken as the target value. Experiment shows that LightGBM (R-2: 0.93) overall performance is better than XGBoost. In some molecules, the predicted fit (R-2) of the coupling constant between atoms even reached 1.00. This research avoids complex quantum mechanics and can assist in NMR to gain insight into the structure and dynamics of molecules, thereby enriching the data information analysis method of nuclear magnetic interaction.