Sleepwalk

Sleepwalk visualizes and assesses the fidelity of dimension-reduced embeddings by mapping inter-point distances onto color gradients to evaluate neighborhood preservation in high-dimensional datasets such as single-cell RNA-seq.

Key Features:

• Interactive visualization: Displays two-dimensional embeddings produced by dimension reduction methods to examine local neighborhood relationships.
• Dynamic coloring mechanism: Colors points according to their distance to a selected focus point, mapping distances to color gradients to reveal neighborhood fidelity.
• Highlighting data characteristics: Uses distance-based color gradients to expose local distortions and hidden structure introduced by dimension reduction.
• Comparative analysis: Supports comparison of multiple embeddings or preprocessing methods to evaluate effects on data representation and multisample integration.

Scientific Applications:

• Single-cell RNA-seq analysis: Assess local structure, neighborhood preservation, and embedding artifacts in single-cell RNA-seq datasets.
• High-dimensional matrix-shaped data: Explore neighborhood relationships and dimension-reduction artifacts in large matrix-form datasets across domains.

Methodology:

Produces 2D embeddings using PCA, MDS, t-SNE, and UMAP; computes distances between a selected focus point and other points and maps those distances to color gradients; enables comparison of different embeddings or preprocessing methods.