Bfimpute

Bfimpute imputes dropout events in single-cell RNA sequencing (scRNA-seq) data using a Bayesian factorization approach to recover latent gene-specific and cell-specific features and improve downstream analyses.

Key Features:

• Bayesian factorization: Employs a Bayesian factorization algorithm to decompose the sparse scRNA-seq count matrix into latent gene-specific and cell-specific feature matrices for imputing dropout events.
• Group-specific imputation: Performs imputation within specified cell groups to tailor recovery of missing values to group-specific expression patterns.
• Integration of additional information: Incorporates user-provided gene- or cell-related information, including cell type labels or bulk RNA-seq data, to refine imputation results.
• Comparative performance: Demonstrated superior performance against six other publicly noted scRNA-seq imputation methods on simulated and real datasets across multiple evaluation metrics.

Scientific Applications:

• Cell-to-cell variation analysis: Improves detection and quantification of cell-to-cell variation by recovering expression values masked by dropouts.
• Gene regulatory network reconstruction: Enables more accurate reconstruction of cell-type-specific gene regulatory networks through recovery of missing expression data.
• Downstream single-cell analyses: Enhances clustering, differential expression analysis, and trajectory inference by providing imputed scRNA-seq expression matrices.

Methodology:

Factorizes the sparse scRNA-seq count matrix into latent gene-specific and cell-specific feature matrices using a Bayesian factorization algorithm that provides probabilistic quantification of uncertainty in imputed values; supports group-specific imputation and incorporation of gene- or cell-level auxiliary information.