Accurate NMR shift prediction using machine learning allows researchers to validate structures by comparing predicted and observed shifts, but predicting 2D NMR remains challenging due to limited annotated data. Here, the authors introduce an unsupervised training framework for predicting and annotating cross-peaks in 2D NMR considering solvent effect, specifically Heteronuclear Single Quantum Coherence (HSQC), showing a MAE of 2.05 ppm for 13C and 0.165 ppm for 1H, and achieving 95.21% concordance with experts’ assignments.
- Yunrui Li
- Hao Xu
- Pengyu Hong