ALRA

ALRA imputes dropout events in single-cell RNA-sequencing (scRNA-seq) data using a low-rank approximation to recover true gene expression while preserving true biological zeros.

Key Features:

• Low-Rank Approximation: ALRA employs a low-rank approximation to impute missing values in scRNA-seq matrices and distinguish dropout (false) zeros from true biological zeros.
• Preservation of Biological Zeros: ALRA maintains true biological zeros at zero to preserve genuine gene silencing or absence in specific cells.
• Validation and Comparison: ALRA was validated against two state-of-the-art imputation techniques and demonstrated superior recovery of marker gene expression while retaining biological zeros.
• Enhanced Cell Type Separation: ALRA improves separation and identification of known cell types within scRNA-seq datasets by refining gene expression profiles.
• Improved Correlation with True Profiles: ALRA increases correlation between imputed cells and their true expression profiles, supporting downstream analyses such as clustering, differential expression, and trajectory inference.
• Scalability: ALRA is scalable to large scRNA-seq datasets, enabling application to extensive single-cell studies.

Scientific Applications:

• Cell Type Identification and Classification: ALRA enhances accuracy of cell type identification and classification by improving gene expression estimates.
• Developmental Biology Studies: ALRA supports analyses of cellular differentiation and development by clarifying gene expression dynamics across cells.
• Cancer Research: ALRA aids investigation of tumor heterogeneity by improving resolution of single-cell expression profiles to identify subclonal populations and trajectories.

Methodology:

Imputation using a low-rank approximation to distinguish dropout versus biological zeros, with validation by comparison to two other state-of-the-art imputation techniques.