NMR shift prediction from small data quantities.

Area of Science:

• Computational Chemistry
• Machine Learning
• Nuclear Magnetic Resonance Spectroscopy

Background:

• Nuclear Magnetic Resonance (NMR) chemical shift prediction typically requires extensive datasets for optimal performance.
• Limited data availability, particularly for heteronuclei, presents a challenge for existing predictive models.

Purpose of the Study:

• To develop and evaluate a novel machine learning model for predicting NMR chemical shifts.
• To demonstrate superior performance of the new model on datasets with limited data compared to existing methods.

Main Methods:

• Development of a novel machine learning architecture tailored for chemical shift prediction.
• Training and validation of the model using datasets with varying amounts of data, focusing on small molecules in specific solvents.
• Comparative analysis against established machine learning models.

Main Results:

• The novel machine learning model achieved higher accuracy in predicting [Formula: see text] and [Formula: see text] NMR chemical shifts compared to other models.
• The model demonstrated significant effectiveness even with comparatively low amounts of data, outperforming benchmarks in these scenarios.

Conclusions:

• The developed machine learning model offers a robust solution for NMR chemical shift prediction, particularly when data is scarce.
• This approach advances the applicability of computational methods in NMR spectroscopy for challenging cases like heteronuclei.