Quantitative Prediction of Exchangeable Proton Chemical Shifts

Area of Science:

• Computational Chemistry
• Spectroscopy
• Machine Learning

Background:

• Predicting NMR chemical shifts of exchangeable protons in solution is difficult due to complex solute-solvent interactions and molecular dynamics.
• Existing methods often struggle to accurately capture these dynamic and interactive effects.

Purpose of the Study:

• To develop a rapid and accurate computational framework for predicting NMR chemical shifts of exchangeable protons in solvated molecules.
• To integrate machine-learning molecular dynamics (ML-MD) with the ShiftML3 shielding model to account for solvation and dynamics.

Main Methods:

• Developed a novel framework combining ML-MD with the ShiftML3 machine-learning shielding model.
• Applied the framework to diverse chemical systems including water, alcohols, nucleobases, glucose, and acetamides.
• Validated predictions against experimental NMR chemical shift data.

Main Results:

• The ML-MD + ShiftML3 framework achieved near-quantitative accuracy in predicting experimental chemical shifts of exchangeable protons.
• The method successfully captured subtle hydrogen-bonding and conformational effects.
• It outperformed implicit-solvent Density Functional Theory (DFT) in accuracy for these systems.

Conclusions:

• ML-MD + ShiftML3 provides a transferable and computationally efficient approach for NMR chemical shift prediction.
• This framework enables realistic spectral predictions for flexible, hydrogen-bonded, and complex molecular systems.
• It advances the capability of NMR spectroscopy by incorporating crucial solvation and dynamic information.