scds: computational annotation of doublets in single-cell RNA sequencing data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution transcriptional heterogeneity studies.
• Doublets, where multiple cells are erroneously identified as single cells, can significantly bias scRNA-seq analysis outcomes.
• Effective computational strategies for doublet identification are essential to ensure reliable biomedical research conclusions.

Purpose of the Study:

• To introduce novel computational methods for in silico doublet identification in scRNA-seq data.
• To provide accurate and efficient tools for detecting and annotating doublets, thereby improving data integrity.
• To develop a scalable solution for doublet annotation in large-scale scRNA-seq datasets.

Main Methods:

• Co-expression based doublet scoring (cxds) utilizes binarized gene expression data and a binomial model for doublet annotation.
• Binary classification based doublet scoring (bcds) employs a binary classification approach to distinguish artificial doublets from real cells.
• The scds package implements both cxds and bcds, allowing for direct application to scRNA-seq datasets.

Main Results:

• The proposed scds methods (cxds and bcds) demonstrate performance comparable to state-of-the-art doublet identification approaches.
• scds achieves this performance with minimal computational cost, making it suitable for large datasets.
• Analysis across four diverse datasets revealed variations in method performance, indicating no single approach universally dominates.

Conclusions:

• The scds package provides a scalable and competitive computational approach for doublet identification in scRNA-seq data.
• scds can annotate datasets with thousands of cells rapidly, in a matter of seconds.
• The developed methods contribute to more robust and reliable downstream analyses of single-cell transcriptional data.