A deep learning model for predicting selected organic molecular spectra.

Area of Science:

• Computational Chemistry
• Machine Learning
• Spectroscopy

Background:

• Accurate molecular spectra simulations are vital for substance discovery and identification.
• Traditional quantum chemistry methods are computationally expensive, limiting efficiency.

Purpose of the Study:

• To develop a deep learning model, DetaNet, for efficient and accurate molecular spectra prediction.
• To overcome the limitations of conventional quantum chemistry calculations.

Main Methods:

• DetaNet combines E(3)-equivariance and self-attention mechanisms.
• The model processes high-order geometric tensorial messages to predict diverse molecular properties.
• Generalized modules were developed for predicting infrared, Raman, UV-Vis, and NMR spectra.

Main Results:

• DetaNet achieves quantum chemistry calculation accuracy for molecular spectra prediction.
• The model demonstrates improved efficiency compared to traditional methods.
• Successfully predicted four key types of molecular spectra on the QM9S dataset.

Conclusions:

• DetaNet offers a significant speed-up for molecular spectra prediction at high accuracy.
• This deep learning approach can facilitate real-time structural identification via spectroscopic measurements.
• Accelerates progress in computational chemistry and materials science.