RCA2: a scalable supervised clustering algorithm that reduces batch effects in scRNA-seq data.

Area of Science:

• Genomics
• Computational Biology
• Immunology

Background:

• Single-cell (SC) technologies offer detailed analysis of human cell transcriptomic diversity.
• Unsupervised clustering, the standard for cell type definition, often groups cells by technical rather than biological variation.
• Existing methods are susceptible to batch effects and data quality issues, limiting accuracy.

Purpose of the Study:

• To present RCA2, a novel algorithm for scalable and robust supervised clustering of single-cell transcriptomes.
• To improve cell type identification by mitigating technical variations and data artifacts.
• To provide a user-friendly framework with integrated analysis modules and reference panels.

Main Methods:

• Developed RCA2, combining reference projection for batch effect robustness with graph-based clustering for scalability.
• Incorporated cell type-specific quality control (QC) for heterogeneous tissues.
• Utilized benchmark datasets and real-world SC data from human bone marrow, PBMCs, and COVID-19 patient PBMCs.

Main Results:

• Demonstrated that supervised clustering with RCA2 is robust to batch effects and data quality artifacts, outperforming unsupervised clustering.
• RCA2 successfully identified cell types in complex SC datasets, including those from heterogeneous tissues.
• The algorithm provides scalable analysis suitable for cohort-scale single-cell datasets.

Conclusions:

• RCA2 offers a robust and scalable solution for supervised clustering in single-cell transcriptomics.
• The method enhances cell type identification accuracy by addressing technical variations.
• RCA2 facilitates unified analysis of large-scale single-cell datasets, advancing biological discovery.