Comprehensive and Explainable Fragmentation: A Machine Learning Approach for Fast and Accurate Mass Spectrum...

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Machine Learning

Background:

• Mass spectrometry (MS) is crucial for chemical identification but current in-silico prediction tools have limitations.
• Existing methods struggle with broad instrument conditions, large molecular libraries, and detailed fragment structure analysis.

Purpose of the Study:

• To develop an advanced in-silico prediction strategy for mass spectrometry.
• To overcome limitations in predicting spectral properties and identifying molecular fragments.
• To enhance the understanding of molecular fragmentation behavior.

Main Methods:

• Proposed a dual-model machine learning strategy combining classification and regression.
• Utilized a classification model for fragment identification and noise filtering.
• Employed a regression model for accurate spectral prediction, incorporating an attention mechanism.

Main Results:

• The dual-model approach demonstrated superior accuracy and efficiency compared to existing algorithms.
• The attention mechanism significantly improved the prediction of molecular fragmentation patterns.
• Successfully facilitated large-scale in-silico spectra calculations.

Conclusions:

• The developed machine learning strategy offers a robust solution for in-silico mass spectrometry.
• It enables deeper insights into molecular fragmentation and aids in analyzing unknown structures.
• Promotes broader applications of MS in chemical identification and analysis.