Diffusive topology preserving manifold distances for single-cell data analysis.

Area of Science:

• Computational Biology
• Bioinformatics
• Data Science

Background:

• High-dimensional single-cell data analysis is crucial for biological discovery.
• Existing dimensionality reduction methods often struggle with preserving true data relationships due to 2D visualization limitations.

Purpose of the Study:

• To introduce Diffusive Topology Neighbor Embedding (DTNE), a novel dimensionality reduction framework.
• To enhance the accurate approximation of manifold distances for improved cellular relationship and dynamics analysis.

Main Methods:

• DTNE utilizes a modified personalized PageRank algorithm to construct a manifold distance matrix.
• This approach preserves the topological structure of high-dimensional data.
• The framework supports integrated analyses including cellular relationship analysis, pseudotime inference, and clustering.

Main Results:

• DTNE demonstrates superior performance in maintaining geodesic distances compared to mainstream algorithms.
• Benchmarking on diverse datasets confirms DTNE's effectiveness in revealing significant biological patterns.
• The method successfully facilitates distribution-based cellular relationship analysis and pseudotime inference.

Conclusions:

• DTNE is a powerful tool for high-dimensional data analysis, offering enhanced biological insights.
• The framework provides a unified approach for various single-cell data analyses.
• DTNE effectively uncovers meaningful biological patterns by preserving manifold structure.