Predicting ³¹P NMR Shifts in Large-Scale, Heterogeneous Databases by Gas Phase DFT: Impact of Conformer and Solvent...

Area of Science:

• Computational Chemistry
• Spectroscopy
• Data Science

Background:

• The Ilm-NMR-P31 database contains experimental ³¹P NMR data.
• Expanding this dataset with theoretical information can enhance its utility.
• Machine learning (ML) models require comprehensive and accurate data for development.

Purpose of the Study:

• To enrich the Ilm-NMR-P31 dataset with quantum-chemically derived ³¹P chemical shifts and molecular geometries.
• To evaluate the accuracy of computational methods for predicting ³¹P NMR shifts.
• To facilitate the development of advanced ML models for structure elucidation.

Main Methods:

• Density Functional Theory (DFT) calculations at BLYP and B3LYP levels were used.
• Molecular geometries were optimized, and ³¹P chemical shifts were calculated for 10,007 phosphorus compounds.
• Conformational sampling and implicit solvent models were investigated to improve accuracy.

Main Results:

• A hybrid dataset combining experimental and quantum-chemical data was created.
• Root mean squared error (RMSE) for calculated shifts was 30.82 ppm (vacuum) and improved to 29.37 ppm with conformer sampling.
• The B3LYP functional demonstrated sufficient accuracy and broad applicability for routine NMR shift prediction.

Conclusions:

• Quantum-chemical enrichment of NMR data significantly enhances experimental datasets.
• Conformational sampling is crucial for improving the accuracy of predicted NMR shifts.
• The B3LYP functional is a reliable choice for large-scale ³¹P NMR shift prediction, supporting ML applications.