Predicting Discrete Structural Transformations in Small Molecules from Tandem Mass Spectrometry.

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Metabolomics

Background:

• Tandem mass spectrometry (MS/MS) generates spectral data crucial for molecular annotation.
• High-throughput MS/MS data acquisition outpaces manual annotation, creating a bottleneck.
• Existing computational methods like molecular networking offer similarity scores but limited actionable insights for structural elucidation.

Purpose of the Study:

• To develop a novel computational method for quantifying discrete structural transformations between molecules using MS/MS data.
• To improve the accuracy and actionability of metabolite annotation in complex datasets.
• To accelerate the discovery of new molecules from mass spectrometry data.

Main Methods:

• Introduced the Molecular Transformation Graph Edit Measure (MT-GEM) to quantify structural differences via graph edit distance.
• Developed the Spectrum Transformation Edit Predictor (STEP), an ensemble machine learning model, to predict MT-GEM distances from MS/MS spectra.
• Evaluated STEP's performance against state-of-the-art similarity metrics using benchmark datasets and human gut microbial data.

Main Results:

• STEP achieved 48.4% average precision in identifying single structural transformations, a tenfold improvement over existing methods.
• STEP identified three times more single-transformation metabolite pairs than feature-based molecular networking in gut microbial data.
• STEP successfully identified novel drug metabolites and natural product analogs missed by conventional approaches.

Conclusions:

• MT-GEM and STEP provide discrete structural transformation predictions, enabling hypothesis-driven metabolite annotation.
• The developed method significantly enhances the speed and accuracy of molecular discovery from MS/MS data.
• This approach promises to accelerate the identification of new molecules in metabolomics research.