Optimal transport analysis of high-dimensional flow cytometry data in immuno-oncology

Area of Science:

• Computational biology
• Single-cell analysis
• Machine learning

Background:

• Single-cell and spatial profiling generate complex, high-dimensional data.
• Analyzing heterogeneous samples, especially across multiple comparisons (longitudinal, treatment groups, multicenter trials), is challenging.
• Traditional visualization tools like UMAP and t-SNE have limitations due to their stochastic and parameter-sensitive nature.

Purpose of the Study:

• To introduce and apply Optimal Transport (OT) as a robust computational framework for analyzing high-dimensional single-cell data.
• To enable reproducible and quantitative comparisons of cellular state distributions in complex biological samples.
• To overcome the limitations of existing visualization methods for longitudinal and comparative analyses.

Main Methods:

• Optimal Transport (OT), specifically the Sinkhorn algorithm, was employed for computationally tractable analysis of high-dimensional data.
• An OT-based graph representation was constructed using longitudinal data from a clinical trial (NCT06016179).
• The graph encoded population abundance (vertex radii) and inter-population similarity based on OT distance (edge thickness/color).

Main Results:

• The OT-based graph enabled rapid identification of significant population changes, e.g., a decrease in CD8+/IFNɣ+ T cells post-treatment.
• Graph edit distance (GED) quantified overall population and phenotypic shifts across fluorescence parameters.
• This approach provided a reproducible measure of change in high-dimensional cellular state space.

Conclusions:

• Optimal Transport offers a powerful, reproducible method for analyzing cellular heterogeneity and treatment response in complex biological systems.
• This framework integrates with machine learning and can be combined with unbiased clustering for enhanced scalability.
• Future applications include quantifying disease, tracking immune responses, and modeling host-microbiome interactions, providing a comprehensive view of tumor evolution.