Chemical shifts in molecular solids by machine learning

Area of Science:

• Solid-state Nuclear Magnetic Resonance (NMR) Spectroscopy
• Computational Chemistry
• Materials Science

Background:

• NMR chemical shifts are crucial for elucidating solid and amorphous material structures.
• Calculating these shifts typically requires computationally expensive first-principles methods.
• Existing machine learning approaches for solid chemical shifts are limited by vast chemical spaces and data scarcity.

Purpose of the Study:

• To develop a machine learning (ML) method for accurate prediction of NMR chemical shifts in molecular solids.
• To achieve predictive accuracy comparable to Density Functional Theory (DFT) calculations.
• To demonstrate the utility of ML-predicted shifts for experimental structure determination.

Main Methods:

• A novel machine learning model was developed, focusing on local atomic environments.
• The model was trained to predict NMR chemical shifts for molecular solids and their polymorphs.
• The accuracy of ML-predicted shifts was validated against DFT calculations and experimental data.

Main Results:

• The proposed ML method accurately predicts NMR chemical shifts for molecular solids, achieving DFT-level accuracy.
• The model successfully determined the structures of cocaine and a complex pyrazole derivative by matching predicted and experimental shifts.
• The approach overcomes limitations of traditional computational methods and data availability.

Conclusions:

• Machine learning, based on local environments, offers a powerful and accurate alternative for predicting solid-state NMR chemical shifts.
• This method significantly advances the structure elucidation of powdered and amorphous materials.
• The ML model provides a practical tool for identifying unknown solid-state structures, particularly in pharmaceutical research.