scNPF

scNPF applies network propagation and fusion to preprocess single-cell RNA-sequencing (scRNA-seq) data, recovering gene expression and learning cell similarities to improve downstream analyses such as clustering and differential expression.

Key Features:

• Network propagation and fusion: Combines network propagation and fusion techniques to augment scRNA-seq preprocessing.
• Gene expression recovery: Recovers gene expression losses and corrects measurements, including for low and moderately expressed genes.
• Cell similarity learning: Learns similarities between cells and constructs cell similarity networks from scRNA-seq data.
• Integration of interaction networks: Leverages publicly available molecular gene-gene interaction networks as prior knowledge.
• Context-specific topology augmentation: Utilizes context-specific topology inherent in scRNA-seq experiments to augment gene-gene relationships in a data-driven manner.
• Global plus context-specific integration: Integrates global topological information from pre-existing interaction networks with context-specific scRNA-seq data to capture shared and complementary knowledge across datasets.
• Performance benchmarking: Demonstrates comparable or higher accuracy in internal validation metrics and clustering accuracy across diverse scRNA-seq datasets.
• R package implementation: Provided as an R package intended for integration into single-cell RNA-seq analysis pipelines.

Scientific Applications:

• Cell type clustering: Improves clustering accuracy by providing refined cell similarity measures.
• Differential expression analysis: Enhances quantification for low and moderately expressed genes to support more reliable differential expression testing.
• Dimension reduction: Improves preprocessing inputs for dimension reduction methods.
• Visualization: Supports improved visualization of single-cell transcriptomic structure via enhanced preprocessing.
• Cross-dataset knowledge integration: Captures shared and complementary information across diverse scRNA-seq datasets using prior interaction networks.

Methodology:

scNPF combines network propagation and fusion, integrating global topological information from pre-existing molecular gene-gene interaction networks with context-specific topology derived from individual scRNA-seq experiments to recover gene expression values and construct cell similarity networks.