Prediction of spin–spin coupling constants with machine learning in NMR

Abstract

Nuclear magnetic resonance (NMR) spectroscopy is one of the most important methods for analyzing the molecular structures of compounds. The objective in this study is to predict indirect spin-spin coupling constants in NMR based on machine learning. We propose important descriptors for predicting indirect spin-spin coupling constants from target pairs of atoms in molecules, and combine the proposed descriptors with molecular descriptors to predict indirect spin-spin coupling constants with Light-GBM as a regression analysis method. We construct regression models using a dataset and verify their predictive accuracy, and then confirm that the proposed descriptors can predict indirect spin-spin coupling constants more accurately than the traditional descriptors used to predict chemical shifts.

Prediction of chemical shift in NMR: A review

“…The reported error is 0.6-0.8 Hz. Shibata Kaneko [65] use RDKit descriptors and decision trees via LightGBM and achieves errors from 0.880 to 0.99 for 1 H, 2 H, and 3 H HH, CH, and NH couplings. Ito et al [66] use a combination of DFT and ML and report an error of 1.21 Hz for HH couplings.…”