Methods for discovery and characterization of cell subsets in high dimensional mass cytometry data

Area of Science:

• Computational Biology
• Bioinformatics
• Machine Learning

Background:

• Mass cytometry generates high-dimensional single-cell data (>40 features).
• Traditional analysis methods can overlook subtle cell populations.
• Advances in machine learning offer new analytical possibilities.

Purpose of the Study:

• Introduce a computational workflow for high-dimensional mass cytometry data.
• Emphasize unsupervised approaches for cell population identification and characterization.
• Facilitate comprehensive analysis and comparison of biological samples.

Main Methods:

• Sequential application of viSNE, SPADE, and heatmaps.
• Unsupervised machine learning for data visualization and analysis.
• Integration of single-cell and population-level data views.

Main Results:

• Successfully characterized and compared healthy and malignant human tissue samples.
• Identified cell subsets and their unique features.
• Demonstrated the utility of unsupervised methods in revealing unexpected cell phenotypes.

Conclusions:

• The proposed workflow provides a robust framework for mass cytometry data analysis.
• Unsupervised machine learning approaches enhance the discovery of cell populations.
• This methodology supports automated analysis and facilitates future machine learning of cell identity.