Mass spectra prediction with structural motif-based graph neural networks.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Cheminformatics

Background:

• Mass spectrometry is vital for molecular structure identification.
• Spectral library searching is a common analysis method, but limited by database size.
• Expanding spectral databases through prediction is necessary.

Purpose of the Study:

• To propose a novel method for mass spectra prediction.
• To improve the accuracy and scope of spectral databases.
• To address the limitations of current spectral library search methods.

Main Methods:

• Developed the Motif-based Mass Spectrum prediction Network (MoMS-Net), a GNN-based architecture.
• Utilized molecular structural motif information for prediction.
• Represented molecules and substructures in a graph format to capture long-range dependencies efficiently.

Main Results:

• MoMS-Net demonstrated validity across various mass spectra types.
• The model showed superior performance compared to conventional prediction models.
• The GNN approach offered memory efficiency over graph transformers.

Conclusions:

• MoMS-Net effectively predicts mass spectra patterns.
• The proposed method offers a valuable tool for expanding spectral databases.
• This research advances computational approaches in molecular identification.