DynamicViz

DynamicViz generates dynamic visualizations of dimensionality-reduced (DR) projections and quantifies their sensitivity to dataset perturbations using a variance score to improve interpretation of high-dimensional biological data, including single-cell analysis.

Key Features:

• Dynamic visualizations: Produces time- or perturbation-resolved visualizations that show how DR projections change in response to dataset perturbations.
• DR method compatibility: Applies to projections produced by commonly used dimensionality reduction methods (DR) without restriction to a specific algorithm.
• Perturbation sensitivity assessment: Captures the sensitivity of DR visualizations to changes in the data input to reveal potential distortions introduced by projection.
• Variance score metric: Computes a variance score that quantifies the dynamic variability observed within DR visualizations.
• Observation-level variability: Assesses how different observations vary under perturbations to highlight unstable or robust data relationships.
• DR algorithm optimization support: Provides metrics and visual feedback intended to inform optimization and implementation choices for DR algorithms.

Scientific Applications:

• Single-cell analysis: Characterizes variability and reliability of dimensionality-reduced representations in single-cell datasets.
• Dimensionality reduction evaluation: Diagnoses distortions and interpretative pitfalls introduced by DR methods through perturbation-based sensitivity analysis.
• DR method implementation assessment: Informs optimization and validation of DR algorithm implementations using variance-based metrics.

Methodology:

Generate dynamic visualizations of DR projections under dataset perturbations and compute a variance score that quantifies the resulting dynamic variability and sensitivity of observations.