EmbedSOM

EmbedSOM performs dimensionality reduction to produce interpretable embeddings for visualization and analysis of single-cell flow and mass cytometry data.

Key Features:

• Dimensionality Reduction: Performs dimensionality reduction of high-dimensional single-cell flow and mass cytometry measurements, building on principles from FlowSOM.
• Landmark-Directed Embedding Enrichment: Implements a landmark-directed embedding enrichment approach that emphasizes specific regions of interest within the data.
• Compatibility with Manifold-Learning Techniques: Generalizes beyond self-organizing maps (SOMs) to operate with various manifold-learning techniques.
• Inwards-Growing Variant of SOMs: Includes an inwards-growing variant of self-organizing maps to improve embedding structure and interpretability.
• Performance Evaluation and Variants: Has been evaluated against multiple variants that use different landmark-generating functions.

Scientific Applications:

• Cellular Heterogeneity Analysis: Enables identification and exploration of distinct cell subpopulations based on phenotypic marker expression in cytometry datasets.
• Data Integration: Facilitates integration of single-cell cytometry data with other omics datasets for multi-dimensional analyses.
• Biomarker Discovery: Supports identification of candidate biomarkers through improved visualization and interpretation of cytometry-derived cell populations.

Methodology:

EmbedSOM applies a landmark-directed embedding enrichment approach, generalizes to manifold-learning techniques beyond traditional SOMs including an inwards-growing SOM variant, and has been compared across variants using different landmark-generating functions.