CMSSP: A Contrastive Mass Spectra-Structure Pretraining Model for Metabolite Identification

Area of Science:

• Metabolomics
• Computational Chemistry
• Bioinformatics

Background:

• Structural annotation of metabolites from tandem mass spectrometry (MS/MS) data presents a significant challenge in metabolite research.
• Artificial intelligence (AI) is transforming MS data interpretation, aiding in the identification of complex metabolites.
• Current methods often struggle with direct comparison between MS/MS spectra and molecular structures due to distinct data modalities.

Purpose of the Study:

• To introduce CMSSP, a novel Contrastive Mass Spectra-Structure Pretraining framework for enhanced metabolite annotation.
• To create a unified representation space enabling direct comparison of MS/MS spectra and molecular structures.
• To overcome the limitations of analyzing distinct data modalities in metabolite identification.

Main Methods:

• Developed a Contrastive Mass Spectra-Structure Pretraining (CMSSP) framework.
• Transformed MS/MS spectra and molecular structures into a unified modality for similarity-based comparison.
• Evaluated CMSSP on benchmark test sets, including CASMI 2017 and an independent dataset.

Main Results:

• CMSSP significantly improved metabolite annotation accuracy, outperforming state-of-the-art methods.
• Achieved a 30% increase in top-1 accuracy on the CASMI 2017 dataset and a 16% increase in top-10 accuracy on an independent set.
• Demonstrated robust performance across seven chemical categories and high accuracy in analyzing Glycyrrhiza glabra metabolites (86.7% top-1, 100% top-3).

Conclusions:

• CMSSP is a powerful tool for interpreting complex MS/MS data, leading to more accurate and efficient metabolite annotation.
• The framework enhances the analytical capabilities of metabolomics, facilitating deeper understanding of biological systems.
• This approach advances the field by enabling more precise identification of metabolites.