Prediction of Vacuum Ultraviolet/Ultraviolet Gas-Phase Absorption Spectra Using Molecular Feature Representations and...

Area of Science:

• Spectroscopy
• Computational Chemistry
• Machine Learning

Background:

• Ultraviolet (UV) absorption spectroscopy is crucial for analyzing chemical compounds.
• Predicting VUV/UV spectra aids in identifying unknown molecules in various fields.
• Artificial intelligence offers an alternative to quantum chemical spectral prediction.

Purpose of the Study:

• To develop and test machine learning models for predicting gas-phase VUV/UV absorption spectra.
• To introduce and evaluate novel molecular features for enhanced spectral prediction.
• To compare the performance of machine learning models against quantum chemical methods.

Main Methods:

• Utilized structure data files (.sdf) and VUV/UV spectra for 1397 compounds.
• Developed and tested three machine learning models (including Random Forest).
• Introduced new molecular features (ABOCH) to represent pi-bonding, aromaticity, and halogenation.

Main Results:

• The Random Forest regressor achieved the best accuracy with the shortest training time.
• The developed machine learning model outperformed time-dependent density functional theory predictions.
• The new ABOCH features improved spectral prediction accuracy.

Conclusions:

• Machine learning, particularly Random Forest with novel features, provides an efficient and accurate method for VUV/UV spectral prediction.
• This AI-driven approach surpasses traditional quantum chemical methods in speed and accuracy.
• The developed model aids in characterizing volatile and semivolatile chemical compounds.