DP5 without DFT: Uncertainty-calibrated graph neural net accelerates structure confirmation via NMR

Key Points

• The study aims to improve the process of assigning candidate structures to NMR spectra by utilizing a novel method without DFT.
• Evaluated the DP4 and DP5 methods for structure assignment.
• Applied a graph neural network model to enhance uncertainty calibration.
• Analyzed the efficacy of the model in confirming molecular structures based on NMR data.
• The uncertainty-calibrated model outperformed traditional methods in structure confirmation speed.
• Significant reduction in time required for accurate NMR spectrum assignments was observed.
• Potential for broader applications in molecular identification was identified.