scLINE

scLINE learns low-dimensional representations of single-cell RNA-seq (scRNA-seq) data by integrating multiple gene-gene interaction networks through network embedding models to mitigate sparsity from low capture rates and drop-out events while preserving biological signal.

Key Features:

• Network embedding integration: Integrates multiple gene-gene interaction networks using network embedding models to inform representation learning.
• Use of public gene networks: Incorporates existing gene networks sourced from public databases to supplement inter-gene interactions.
• Sparsity and drop-out mitigation: Addresses high sparsity and variability caused by low capture rates and frequent drop-out events in scRNA-seq data.
• Dimensionality reduction with signal preservation: Produces low-dimensional embeddings that preserve essential biological signals from genome-scale transcriptional profiles.
• Comparative performance: Evaluated on eight single-cell datasets and reported to perform comparably or superiorly to PCA, t-SNE, and Isomap based on internal validation metrics and clustering accuracy.
• Downstream utility: Generates embeddings effective for visualization, clustering, and cell typing of single-cell transcriptomic data.
• Implementation: Provided as an R package for integration into scRNA-seq analysis workflows.

Scientific Applications:

• Visualization of transcriptomic landscapes: Facilitates visualization of single-cell transcriptomic heterogeneity in reduced-dimensional space.
• Clustering and cell-type identification: Supports clustering and cell typing based on learned low-dimensional representations.
• Biomedical research domains: Applicable to studies in development, immunology, and cancer that require high-resolution transcriptional profiling.

Methodology:

Applies network embedding models to integrate multiple gene-gene interaction networks sourced from public databases to learn low-dimensional embeddings of scRNA-seq data.