Generalized EmbedSOM on quadtree-structured self-organizing maps.

Area of Science:

• Computational Biology
• Data Science
• Bioinformatics

Background:

• Dimensionality reduction is crucial for analyzing high-dimensional single-cell cytometry data.
• EmbedSOM, a fast algorithm, was initially designed for self-organizing maps in cytometry.
• Existing methods may have limitations in embedding enrichment and data representation.

Purpose of the Study:

• To present an updated, generalized version of EmbedSOM for landmark-directed embedding enrichment.
• To demonstrate its effectiveness with manifold-learning techniques beyond self-organizing maps.
• To introduce an improved inwards-growing self-organizing map variant.

Main Methods:

• Generalization of the EmbedSOM algorithm for broader applicability.
• Integration with various manifold-learning techniques.
• Development of an inwards-growing self-organizing map variant.
• Performance evaluation using different landmark-generating functions.

Main Results:

• The generalized EmbedSOM demonstrates effective embedding enrichment across different manifold-learning methods.
• The inwards-growing self-organizing map variant mitigates previous EmbedSOM output deficiencies.
• Performance comparisons highlight the strengths of various EmbedSOM variants.
• Successful application to real-world single-cell cytometry datasets.

Conclusions:

• The generalized EmbedSOM offers a flexible and powerful tool for single-cell cytometry data analysis.
• The enhanced algorithm improves embedding quality and addresses limitations of earlier versions.
• This work advances dimensionality reduction techniques for complex biological datasets.