scMAGIC: accurately annotating single cells using two rounds of reference-based classification.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data requiring robust cell annotation.
• Existing reference-based classification methods can be limited by batch effects and incomplete reference datasets.

Purpose of the Study:

• To introduce scMAGIC, a novel computational method for accurate cell annotation in scRNA-seq data.
• To improve cell classification by mitigating batch effects and handling incomplete cell type coverage in reference datasets.

Main Methods:

• scMAGIC employs a two-round reference-based classification (RBC) approach.
• Query cells with high confidence classifications in the first round are used as an updated reference for the second round.
• The method can also utilize atlas datasets as references when no specific reference is available.

Main Results:

• scMAGIC significantly outperforms 13 competing RBC methods across 86 benchmark tests.
• The method demonstrates superior performance when query datasets have unique cell types not present in the reference.
• scMAGIC effectively reduces the impact of batch effects between reference and query data.

Conclusions:

• scMAGIC provides a powerful and flexible tool for scRNA-seq data annotation.
• The two-round RBC strategy enhances classification accuracy and robustness.
• This method advances the utility of scRNA-seq analysis in diverse biological contexts.