Unsupervised machine learning for mass spectrometry imaging data analysis with in vivo isotope labeling

Area of Science:

• Metabolomics
• Spatial Biology
• Biophysics

Background:

• Mass spectrometry imaging (MSI) is crucial for spatial metabolomics.
• Untargeted MSI data analysis and MSI with in vivo isotope labeling (MSIi) present significant challenges.
• Machine learning (ML) applications for MSIi analysis remain unexplored.

Purpose of the Study:

• To explore unsupervised machine learning applications for MSIi data analysis.
• To leverage ML for differentiating metabolite localizations and investigating isotope labeling in untargeted metabolites.
• To assess the impact of ML on analysis time, throughput, and clarity of spatial isotopologue distributions.

Main Methods:

• Utilized the Cardinal software to process MSIi datasets from ¹³CO₂ and D₂O labeled duckweed.
• Applied spatial shrunken centroid (SSC) segmentation, an unsupervised ML algorithm, for data analysis.
• Calculated the fraction of de novo biosynthesis to gain insights into tissue-specific metabolite flux.

Main Results:

• SSC segmentation identified five distinct spatial segments in ¹³C-labeled duckweed based on lipid isotopologue distributions, surpassing manual classification.
• SSC segmentation of D-labeled duckweed revealed five spatial segments characterized by unique metabolite and isotopologue profiles.
• Untargeted segmentation analysis provided insights into tissue-specific relative flux of metabolites.

Conclusions:

• Unsupervised machine learning significantly enhances the analysis of complex MSIi datasets.
• ML application reduces analysis time and increases throughput for spatial metabolomics.
• This approach improves the clarity and depth of understanding spatial isotopologue distributions and metabolic dynamics.