scDA

scDA identifies cell groups and discriminant metagenes from single-cell RNA sequencing (scRNA-seq) data to annotate unlabeled cells and characterize cellular heterogeneity across large and multi-batch datasets.

Key Features:

• Cell-by-cell representation graph: Constructs a cell-by-cell representation graph to capture relationships among single cells.
• Discriminant metagenes: Identifies discriminant metagenes that characterize molecular signatures of cell groups.
• Annotation of unlabeled cells: Uses identified cell groups and discriminant metagenes to annotate unlabeled cells within datasets.
• Cell type determination across datasets: Determines cell types and reveals overall variabilities between cells from eleven different data sets.
• Performance relative to existing methods: Outperforms several state-of-the-art methods when inferring labels of new samples.
• Robustness to drop-out events: Exhibits reduced sensitivity to drop-out events common in scRNA-seq data.
• Multi-batch and large-scale handling: Efficiently analyzes large-scale or multi-batch scRNA-seq profiles and can label large numbers of cells when trained on a small set.

Scientific Applications:

• Exploration of cellular heterogeneity: Characterizes phenotypic and molecular heterogeneity within complex biological systems using scRNA-seq data.
• Developmental biology: Identifies and annotates cell types and states across developmental trajectories.
• Cancer research: Resolves tumor cellular composition and heterogeneity by annotating malignant and non-malignant cell populations.
• Immunology: Profiles immune cell diversity and identifies discriminant gene signatures across immune cell types.

Methodology:

Constructs a cell-by-cell representation graph, identifies discriminant metagenes from that graph, and uses those metagenes to annotate unlabeled cells.