Machine-Learning Approach to Identify Organic Functional Groups from FT-IR and NMR Spectral Data

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Spectroscopy

Background:

• Chemical structure elucidation from spectral data is time-consuming.
• Traditional methods often rely on single spectroscopic techniques, limiting accuracy.
• Advancements in machine learning offer potential for improved data analysis.

Purpose of the Study:

• To develop a machine-learning model for rapid and accurate identification of functional groups in unknown compounds.
• To evaluate the performance of a multi-spectroscopic data approach compared to single-technique models.
• To enhance the efficiency of chemical structure analysis.

Main Methods:

• Developed an artificial neural network model.
• Trained the model on combined Fourier-transform infrared (FTIR), proton nuclear magnetic resonance (¹H NMR), and carbon-13 nuclear magnetic resonance (¹³C NMR) spectral data.
• Evaluated model performance using macro-average F1 score for functional group identification.

Main Results:

• The multi-spectroscopic machine-learning model successfully identified 17 functional groups.
• Achieved a macro-average F1 score of 0.93, demonstrating high accuracy.
• Outperformed models trained on single types of spectroscopic data.

Conclusions:

• Integrating multiple spectroscopic data sources into machine-learning models significantly improves the accuracy and speed of functional group identification.
• This approach offers a more robust method for analyzing the structure of unknown chemicals.
• Simultaneous use of multiple spectroscopy methods, powered by AI, enhances chemical analysis.