SIEVE: identifying robust single cell variable genes for single-cell RNA sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis involves multiple steps, including quality control, normalization, and highly variable gene selection.
• Downstream analyses like dimensionality reduction and clustering are highly sensitive to the choice of variable genes.
• Existing tools for variable gene selection lack comprehensive performance evaluations and a standardized strategy.

Purpose of the Study:

• To evaluate the performance of nine common methods for screening variable genes in scRNA-seq data.
• To propose a new, robust strategy for identifying variable genes that minimizes stochastic noise.
• To improve the accuracy of single-cell classification by enhancing variable gene identification.

Main Methods:

• Comparative analysis of nine variable gene selection methods using scRNA-seq data from hematopoietic stem/progenitor and mature blood cells.
• Development of the SIEVE (SIngle-cEll Variable gEnes) strategy employing multiple rounds of random sampling.
• Assessment of SIEVE's ability to identify robust variable gene sets, including lowly expressed genes.

Main Results:

• The SCHS method showed high reproducibility and accuracy but favored highly expressed genes.
• The proposed SIEVE strategy effectively minimizes stochastic noise and identifies a robust set of variable genes.
• SIEVE successfully recovers lowly expressed genes and significantly improves single-cell classification accuracy, particularly for methods with lower reproducibility.

Conclusions:

• SIEVE offers a superior approach for variable gene selection in scRNA-seq data, addressing limitations of existing methods.
• The SIEVE strategy enhances the reliability and accuracy of downstream analyses, including cell classification.
• The SIEVE software is publicly available to facilitate its adoption in scRNA-seq data analysis.