Novel machine learning approach to differential cell flow cytometry analysis based on projection pursuit

Area of Science:

• Computational Biology
• Biotechnology
• Data Science

Background:

• Multicolor cell flow cytometry is a key technique for identifying cell subpopulations based on physical and biochemical properties.
• Current flow cytometry analysis relies on manual gating, which can be inconsistent, arbitrary, and may miss complex data structures.
• Existing methods often analyze data in two dimensions, potentially masking higher-dimensional patterns.

Purpose of the Study:

• Introduce differential projection pursuit (DPP) as a novel methodology for analyzing large datasets.
• Apply DPP to cell flow cytometry data as an alternative to traditional gating methods.
• Enhance the clarity, reproducibility, and automation of flow cytometry data analysis.

Main Methods:

• Developed differential projection pursuit (DPP), integrating projection pursuit, data nuggets, and factor analysis.
• Applied DPP to a multicolor cell flow cytometry dataset.
• DPP identifies regions with maximal differences between experimental treatments or distributions.

Main Results:

• DPP successfully identified differences in cell populations under varying experimental conditions.
• The methodology demonstrated potential for automating flow cytometry analysis.
• DPP considers data in its true dimensional space, overcoming limitations of 2D plotting.

Conclusions:

• Differential projection pursuit provides an automated and more robust alternative to manual gating in flow cytometry.
• This method improves the identification of cell subpopulations and enhances analytical reproducibility.
• DPP offers a platform for exploring differences in large datasets and advancing flow cytometry analysis.