Leveraging multiple labeled datasets for the automated annotation of single-cell RNA and ATAC data

Area of Science:

• Single-cell genomics
• Computational biology
• Bioinformatics

Background:

• Single-cell atlases are crucial for understanding cellular diversity but face challenges from batch effects and accurate cell annotation.
• Current single-cell RNA sequencing (scRNA-Seq) and single-cell ATAC sequencing (scATAC-Seq) integration methods are not optimized for cell annotation.
• Existing annotation tools often depend on external databases or reference datasets, limiting adaptability for specific needs, especially for rare cell types or scATAC-Seq data.

Purpose of the Study:

• To introduce JIND-Multi, a novel framework for transferring cell-type labels across multiple annotated single-cell datasets.
• To develop a versatile annotation tool applicable to both scRNA-Seq and scATAC-Seq data.
• To improve the accuracy and reduce the proportion of unclassified cells in single-cell atlases.

Main Methods:

• JIND-Multi utilizes a label transfer approach across multiple annotated datasets.
• The framework is designed to work with scRNA-Seq data using annotated scRNA-Seq data.
• It also handles scATAC-Seq data using annotated scATAC-Seq data, unlike methods requiring paired scRNA-Seq data.

Main Results:

• JIND-Multi significantly reduces the proportion of unclassified cells in both scRNA-Seq and scATAC-Seq datasets.
• The framework maintains the accuracy and performance of the original JIND model.
• JIND-Multi demonstrates favorable comparisons against state-of-the-art annotation methods.

Conclusions:

• JIND-Multi offers a versatile and effective solution for cell annotation across different single-cell sequencing technologies.
• The framework improves annotation precision by leveraging multiple labeled datasets, reducing unassigned cells.
• JIND-Multi provides a reliable tool for the single-cell research community, enhancing the creation of comprehensive cell atlases.