Fully Automated Unconstrained Analysis of High-Resolution Mass Spectrometry Data with Machine Learning

Area of Science:

• Analytical Chemistry
• Computational Chemistry
• Biochemistry

Background:

• Mass spectrometry (MS) is crucial for analyzing complex mixtures in materials science, life sciences (metabolomics, proteomics), and chemistry.
• Complete signal assignment in MS data remains a significant challenge, hindering comprehensive analysis.
• A "dream tool" for researchers would be an algorithm capable of unconstrained formula generation and component identification from spectra.

Purpose of the Study:

• To present a framework for efficient mass spectrometry (MS) data interpretation.
• To introduce a novel approach for detailed MS analysis by tackling the inverse spectral problem.
• To develop a method for generating molecular formulas from isotopic structures.

Main Methods:

• Deisotoping using gradient-boosted decision trees.
• A neural network for generating molecular formulas from fine isotopic structures.
• Framework tested on proteomics (protein sequencing), life sciences (natural samples), and chemistry (catalysis).

Main Results:

• Successfully demonstrated a novel approach for MS data interpretation.
• The method effectively generates molecular formulas from isotopic fine structures.
• Validated the framework across diverse applications in proteomics, life sciences, and chemistry.

Conclusions:

• The presented framework offers an efficient solution for complex MS data interpretation.
• The novel approach, combining deisotoping and neural networks, advances the solution to the inverse spectral problem.
• This tool has broad applicability and potential to significantly aid researchers in various scientific disciplines.