scMAGIC

scMAGIC classifies cell types in single-cell RNA sequencing (scRNA-seq) data using marker gene identification and two rounds of reference-based classification to improve annotation accuracy and mitigate batch effects.

Key Features:

• Marker gene identification: Identifies marker genes to guide cell type assignment in query scRNA-seq datasets.
• Two-round reference-based classification (RBC): Performs an initial RBC followed by a second RBC to refine assignments.
• Incorporation of confident query cells: Integrates confidently classified query cells from the first round into the reference pool for the second round.
• Batch-effect mitigation: Uses the iterative incorporation of query cells to mitigate batch effects between reference and query datasets.
• Reference usage: Leverages well-annotated scRNA-seq reference datasets for classification.
• Performance benchmarking: Demonstrated superior performance relative to 13 competing RBC methods across 86 benchmark tests.
• Robustness to incomplete references: Maintains accuracy when cell types in the query dataset are not fully represented in the reference.
• Atlas-as-surrogate capability: Retains high annotation accuracy when no direct reference is available by using an atlas dataset as a surrogate reference.

Scientific Applications:

• Cell type annotation in scRNA-seq studies: Assigns cell types in single-cell transcriptomic datasets for downstream biological interpretation.
• Cross-dataset classification and integration: Enables classification across datasets with batch effects by leveraging iterative reference augmentation.
• Annotation with surrogate references: Supports annotation when direct reference datasets are absent by using atlas datasets as surrogates.
• Method benchmarking and comparison: Serves as a reference method for comparative evaluation against other RBC approaches.
• Studies of cellular heterogeneity and function: Facilitates analyses of cellular composition and functional heterogeneity in complex tissues.

Methodology:

scMAGIC identifies marker genes, performs an initial reference-based classification using a well-annotated scRNA-seq reference, then performs a second RBC after incorporating confidently classified query cells into the reference to reduce batch effects.