Single-Cell Classification Using Mass Spectrometry through Interpretable Machine Learning.

Area of Science:

• Neuroscience
• Computational Biology
• Analytical Chemistry

Background:

• Understanding brain function requires characterizing cellular heterogeneity.
• The relationship between cell chemistry, function, and phenotype remains unclear.
• Mass spectrometry enables detailed chemical analysis of individual cells.

Purpose of the Study:

• To develop and validate a machine learning workflow for classifying single cells based on mass spectrometry data.
• To leverage interpretable AI for understanding chemical distinctions between cell types.
• To identify key chemical features that define specific cell groups of interest.

Main Methods:

• Utilized matrix-assisted laser desorption/ionization mass spectrometry for high-throughput single-cell analysis.
• Developed a machine learning model for classifying single cells from mass spectra.
• Applied SHapley Additive exPlanations (SHAP) for interpretable model analysis and feature importance ranking.

Main Results:

• Achieved >80% classification accuracy for single cells across different platforms and cell types.
• Identified specific mass features crucial for discriminating between neuronal and glial cells.
• SHAP analysis provided both local and global interpretations of chemical profiles, preserving heterogeneity.

Conclusions:

• The developed interpretable machine learning workflow effectively discriminates chemical profiles of single cells.
• This approach facilitates the analysis and validation of cell-specific chemical signatures.
• Provides valuable insights into the chemical heterogeneity underlying cell function in the brain.