Prediction of Natural Product Classes Using Machine Learning and ¹³C NMR Spectroscopic Data.

Area of Science:

• Computational Chemistry
• Natural Product Chemistry
• Machine Learning Applications

Background:

• Structure elucidation of chemical compounds is critical in chemistry.
• Carbon-13 Nuclear Magnetic Resonance (¹³C NMR) spectroscopy provides extensive structural information.
• Natural products (NPs) can be categorized into classes based on structural similarities.

Purpose of the Study:

• To investigate the potential of using ¹³C NMR data for predicting natural product classes.
• To develop and evaluate machine learning models for automated NP classification.

Main Methods:

• Utilized freely available ¹³C NMR data from natural products.
• Trained four distinct machine learning classifiers to predict eight common NP classes.
• Evaluated classifier performance using metrics such as f1-scores, including variations with sample percentages and glycoside presence.

Main Results:

• The XGBoost classifier achieved the highest performance, with f1-scores exceeding 0.82 for NP class prediction.
• Models demonstrated robust performance on data outside the training set, with accuracies above 80% for most classes.
• Specific subclass prediction (coumarins within chromans) resulted in perfect accuracy (100%).

Conclusions:

• ¹³C NMR data, when combined with machine learning, is a powerful tool for predicting natural product classes.
• The XGBoost algorithm shows significant promise for automated classification of natural products.
• This approach can streamline the process of structure elucidation and natural product research.