TopoMetry systematically learns and evaluates the latent geometry of single-cell data.

Area of Science:

• Computational Biology
• Single-Cell Analysis
• Data Geometry

Background:

• Reconstructing data geometry is crucial for single-cell analysis.
• Existing frameworks lack unified methods for learning, evaluating, and diagnosing representations that preserve geometry.

Purpose of the Study:

• Introduce TopoMetry, a geometry-aware framework for single-cell data analysis.
• Develop a method to learn, evaluate, and diagnose data representations that faithfully preserve geometry.

Main Methods:

• TopoMetry learns intrinsic coordinate systems directly from data.
• It refines these coordinates into high-fidelity spectral scaffolds.
• Scaffolds capture both local neighborhoods and global data structures.

Main Results:

• TopoMetry reliably preserved geometry across diverse single-cell datasets.
• It outperformed standard workflows in preserving geometric structures.
• Revealed obscured biological signals, including T cell diversity and clonal expansion.

Conclusions:

• TopoMetry offers a powerful and accessible framework for single-cell analysis.
• It enables more accurate exploration of cellular diversity by focusing on geometry.
• The framework facilitates improved clustering, visualization, and discovery of biological insights.