scTSSR

scTSSR recovers gene expression levels in single-cell RNA sequencing (scRNA-seq) data by imputing dropout events using a two-side sparse self-representation approach.

Key Features:

• Two-Side Sparse Self-Representation Model: Simultaneously leverages similarity among genes and among cells to construct a sparse self-representation for imputation of scRNA-seq data.
• Dropout Imputation: Targets technical dropout events to recover true gene expression values and reduce noise in single-cell datasets.
• Gene Expression Recovery Performance: Demonstrates superior recovery in down-sampling experiments, capturing Gini coefficients and gene-to-gene correlations observed in smRNA FISH.
• Differential Expression Analysis Support: Produces imputed expression matrices that can be used to identify differentially expressed genes across cell populations.
• Cell Clustering Support: Improves grouping of cells by enhancing expression signal prior to clustering analyses.
• Cell Trajectory Inference Support: Supports reconstruction of developmental trajectories by providing denoised expression profiles for trajectory methods.
• Implementation and Dependencies: Implemented in R and relies on SAVER, keras, and tensorflow.

Scientific Applications:

• Developmental Biology: Enables more accurate reconstruction of developmental trajectories and gene expression programs at single-cell resolution.
• Cancer Research: Facilitates analysis of tumor heterogeneity by recovering expression signals lost to scRNA-seq dropouts.
• Immunology: Improves detection of immune cell subpopulations and expression patterns from scRNA-seq data.
• Studies of Cellular Heterogeneity and Complex Systems: Enhances quantitative analysis of cell-to-cell variability and gene–gene relationships in systems relying on scRNA-seq.

Methodology:

Uses a two-side sparse self-representation model combining information from similar genes and similar cells for imputation; evaluated via down-sampling experiments and comparison to smRNA FISH using Gini coefficients and gene-to-gene correlations; implemented in R with dependencies on SAVER, keras, and tensorflow.