SIEVE

SIEVE identifies robust variable genes from single-cell RNA sequencing (scRNA-seq) data to improve downstream dimensionality reduction and cell classification.

Key Features:

• Robust Variable Gene Identification: Uses multiple rounds of random sampling to minimize stochastic noise and identify a reliable set of highly variable genes.
• Recovery of Lowly Expressed Genes: Recovers lowly expressed genes as variable genes to support more accurate single-cell classification.
• Comparative Performance Evaluation: Benchmarked against nine commonly used methods for screening variable genes using scRNA-seq data from hematopoietic stem/progenitor cells and mature blood cells.
• Enhanced Reproducibility and Accuracy: Demonstrates improved reproducibility and accuracy compared with other variable-gene selection strategies.
• Addresses scRNA-seq QC and Normalization Challenges: Mitigates effects of quality control and normalization issues on the selection of highly variable genes and subsequent analyses.

Scientific Applications:

• Hematopoiesis studies: Identification of variable genes in hematopoietic stem/progenitor cells and mature blood cells for cell-state and lineage analyses.
• Single-cell classification: Improves accuracy of classifying cell types and states in scRNA-seq datasets.
• Dimensionality reduction and clustering: Provides a more reliable variable-gene set for PCA, t-SNE, UMAP, and clustering workflows.
• Detection of low-abundance signals: Enhances identification of biologically relevant, lowly expressed genes in single-cell analyses.

Methodology:

SIEVE applies multiple rounds of random sampling to mitigate stochastic noise in variable gene identification and was benchmarked against nine commonly used variable-gene screening methods using scRNA-seq data from hematopoietic stem/progenitor cells and mature blood cells.