Feature selection methods affect the performance of scRNA-seq data integration and querying.

Area of Science:

• Single-cell transcriptomics
• Computational biology
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables reference cell atlas construction.
• Effective dataset integration and new sample mapping are crucial for atlas utility.
• Previous benchmarks highlight feature selection's importance but lack detailed guidance.

Purpose of the Study:

• To benchmark various feature selection methods for scRNA-seq data integration.
• To evaluate methods beyond standard batch correction and biological variation preservation.
• To assess performance in query mapping, label transfer, and novel population detection.

Main Methods:

• Benchmarking feature selection strategies for scRNA-seq integration.
• Utilizing metrics for query mapping, label transfer, and unseen population detection.
• Analyzing the impact of feature number, batch-awareness, and lineage-specificity.

Main Results:

• Highly variable feature selection is confirmed as an effective strategy for high-quality integration.
• Provides guidance on the optimal number of features to select.
• Demonstrates the influence of batch-aware and lineage-specific feature selection.

Conclusions:

• Reinforces the utility of highly variable feature selection for scRNA-seq data integration.
• Offers practical recommendations for feature selection in large-scale atlas construction.
• Informs analysts on optimizing data integration for biological discovery.