Deep Learning-Assisted Spectrum-Structure Correlation: State-of-the-Art and Perspectives

Area of Science:

• Computational Chemistry
• Spectroscopy
• Machine Learning

Background:

• Spectrum-structure correlation is vital in spectral analysis, with recent advancements driven by high-throughput detection and expanding chemical spaces.
• Traditional chemometrics struggles with the massive data generated, necessitating advanced analytical approaches.

Purpose of the Study:

• To review the application of deep learning in establishing spectrum-structure correlations.
• To explore deep learning's role in spectral prediction (forward) and inverse spectrum-structure correlation (library matching, de novo generation).

Main Methods:

• Introduction to molecular/spectral representations and deep learning fundamentals.
• Summary of deep learning methodologies applied to spectrum-structure correlation over the last five years.

Main Results:

• Deep learning significantly enhances spectral prediction accuracy.
• Deep learning facilitates inverse spectrum-structure correlation, aiding in library matching and novel molecular design.

Conclusions:

• Deep learning offers a powerful approach to overcome limitations in conventional chemometrics for spectrum-structure correlation.
• Continued development in deep learning promises to unlock ultimate solutions for spectrum-structure correlation, driving advancements across scientific disciplines.