Visualizing structure and transitions in high-dimensional biological data.

Area of Science:

• Computational Biology
• Data Visualization
• Bioinformatics

Background:

• High-throughput technologies generate complex, high-dimensional data.
• Existing visualization tools struggle to intuitively represent intricate data structures and patterns.
• Effective visualization is crucial for extracting meaningful insights from biological datasets.

Purpose of the Study:

• To introduce PHATE (Potential of Hydrogen, Attenuated Total Reflectance, and Environmental Analysis), a new visualization method.
• To demonstrate PHATE's ability to capture both local and global nonlinear data structures.
• To compare PHATE's performance against existing visualization techniques.

Main Methods:

• PHATE utilizes an information-geometric distance to map data points.
• The study employed artificial and diverse biological datasets for comparative analysis.
• A novel metric, denoised embedding manifold preservation (DEMaP), was defined and used for quantitative evaluation.

Main Results:

• PHATE consistently preserves diverse data patterns, including progressions, branches, and clusters, superior to other methods.
• PHATE generates quantitatively better denoised, lower-dimensional embeddings compared to existing visualization techniques.
• Analysis of single-cell RNA sequencing data revealed novel biological insights into human germ-layer differentiation, identifying three new subpopulations.

Conclusions:

• PHATE is a powerful visualization tool for high-dimensional biological data.
• It excels at preserving complex data structures and revealing biological insights.
• PHATE demonstrates broad applicability across various data types, including scRNA-seq, mass cytometry, Hi-C, and microbiome data.