scMRA

scMRA applies deep learning to annotate single-cell RNA sequencing (scRNA-seq) data by integrating multiple reference datasets into a meta-dataset and transferring cell-type labels while mitigating batch effects for consistent cross-dataset annotation.

Key Features:

• Deep learning-based annotation: Uses deep learning to perform cell-type annotation of scRNA-seq data.
• Meta-dataset aggregation: Aggregates multiple reference scRNA-seq datasets into a meta-dataset to enable multi-reference annotation.
• Knowledge graph representation: Constructs a knowledge graph that captures characteristics of cell types across datasets.
• Graph convolutional network discriminator: Employs a graph convolutional network as a discriminator to preserve intra-cell-type closeness and maintain relative positioning of cell types across datasets.
• Multi-reference knowledge transfer: Transfers knowledge from multiple reference datasets to annotate unlabeled target domains.
• Batch effect mitigation: Addresses batch effects arising from different sequencing platforms to improve annotation consistency.

Scientific Applications:

• Multi-reference scRNA-seq annotation: Annotating scRNA-seq datasets using multiple, potentially insufficient, reference datasets.
• Cross-dataset cell-type harmonization: Establishing consistent cell-type labels and relative positioning across heterogeneous datasets.
• Transfer learning for unlabeled data: Transferring cell-type labels from reference datasets to unlabeled target domains.
• Batch effect correction in multi-platform studies: Mitigating sequencing-platform–related batch effects during annotation.

Methodology:

Aggregates multiple reference datasets into a meta-dataset, constructs a knowledge graph of cell-type characteristics, and trains a graph convolutional network discriminator to transfer labels to unlabeled target data while mitigating batch effects.