A Hashing-Based Framework for Enhancing Cluster Delineation of High-Dimensional Single-Cell Profiles

Area of Science:

• Computational Biology
• Bioinformatics
• Single-cell Omics

Background:

• Clustering high-dimensional (HD) single-cell omics data is crucial for understanding cell function.
• Inconspicuously differential gene or protein expressions can hinder accurate cell cluster delineation and downstream analysis.
• Existing methods struggle with subtle expression variations that obscure true cellular heterogeneity.

Purpose of the Study:

• To develop a novel hashing-based framework to enhance cell cluster delineation in HD single-cell omics data.
• To improve the utilization of cell biomarkers and uncover hidden cellular heterogeneities.
• To provide a method that complements existing clustering algorithms like PhenoGraph.

Main Methods:

• A hashing-based framework is proposed, projecting data into a sparse HD space.
• Fly and densefly hashing preprocessing techniques are employed to retain data's local structure.
• The framework is designed to decompose non-significant variables into latent variables for improved cell distinction.

Main Results:

• The hashing framework successfully improves the cluster delineation capabilities of existing methods, such as PhenoGraph.
• Analysis of mass cytometry datasets revealed new, previously hidden heterogeneities within cell clusters.
• The method demonstrated enhanced utilization of cell biomarkers, leading to enriched biological findings.

Conclusions:

• The proposed hashing-based framework offers a significant advancement in analyzing HD single-cell omics data.
• It effectively addresses challenges posed by subtle biomarker expression differences, improving cell cluster resolution.
• This approach enriches biological discoveries by uncovering latent variables and cellular subpopulations.