SAKURA: a knowledge-guided approach to recovering important, rare signals from single-cell data

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Dimensionality reduction is crucial for analyzing single-cell transcriptomic data.
• Current methods often overlook rare but significant biological signals.
• Identifying subtle cell differences is vital for understanding complex biological systems.

Purpose of the Study:

• To introduce SAKURA, a novel framework for knowledge-guided dimensionality reduction.
• To enhance the detection and separation of rare and similar cell subpopulations.
• To improve the interpretability of single-cell transcriptomic data.

Main Methods:

• Developed a novel framework named SAKURA.
• Employed knowledge-derived genes of interest to guide the dimensionality reduction process.
• Applied the framework to identify endocrine cell subtypes, hematopoietic subpopulations, and senescent cells.

Main Results:

• SAKURA effectively clusters rare cells and separates highly similar cell subpopulations.
• Demonstrated utility in identifying specific cell subtypes within the pancreatic islet.
• Successfully identified rare senescent cells and distinct hematopoietic subpopulations.

Conclusions:

• SAKURA offers a powerful approach for uncovering hidden cellular heterogeneity.
• Knowledge-guided dimensionality reduction can overcome limitations of existing methods.
• This framework has broad applications in single-cell data analysis for biological discovery.