Advancing Structure Elucidation with a Flexible Multi-Spectral AI Model.

Area of Science:

• Computational chemistry
• Machine learning in chemistry
• Spectroscopic data analysis

Background:

• Chemical synthesis validation relies on analytical techniques, but spectral data interpretation is a bottleneck.
• Automated data collection in high-throughput synthesis exacerbates the need for efficient interpretation methods.

Purpose of the Study:

• To develop an automated machine learning method for predicting chemical structures from multiple spectral data types.
• To address challenges in interpreting spectral data for chemical synthesis validation.

Main Methods:

• Introduction of the MultiModalSpectralTransformer (MMST), a machine learning model.
• Training MMST on 4 million simulated compounds across NMR, IR, and MS spectral data.
• Implementation of an active learning cycle to improve model adaptability to novel chemical structures.

Main Results:

• MMST achieved 72% top-1 and 80% top-3 accuracy in predicting chemical structures.
• The model demonstrated good performance with experimental spectra, despite being trained on simulated data.
• Benchmarking confirmed MMST's capabilities across diverse molecular weight ranges and chemical spaces.

Conclusions:

• MMST represents a significant advancement in automated structure elucidation.
• The method provides a powerful and adaptable tool for bridging simulated and real-world spectral data.
• MMST offers a potential solution to the intensifying challenge of spectral data interpretation in chemical synthesis.