CDSImpute

CDSImpute imputes dropout events in single-cell RNA sequencing (scRNA-seq) data to recover true gene expression profiles and improve downstream analyses such as differential expression and clustering.

Key Features:

• Dropout imputation: Detects and corrects technical dropout events characterized by excessive zero and near-zero counts in scRNA-seq data.
• Correlation and negative distance metrics: Leverages both correlation and negative distance metrics between cells to identify similarity.
• Similar-cell borrowing: Constructs a list of similar cells and borrows their gene expression values to impute missing or dropout measurements.
• Distinction of zeros: Accounts for the difference between technical dropouts and biological zeros when recovering expression profiles.
• Benchmark datasets: Validated on simulation datasets and public scRNA-seq datasets including Kolod, Pollen, and Usoskin.
• Quantitative evaluation: Performance quantified using the adjusted rand index (ARI), reporting ARI values of 1.00 (Kolod), 0.79 (Pollen), and 0.34 (Usoskin).
• Comparative performance: Demonstrates improved clustering accuracy and differential expression detection relative to three other evaluated methods.

Scientific Applications:

• Expression recovery: Recovers true gene expression profiles in scRNA-seq datasets by imputing technical dropouts.
• Cell-type identification: Improves clustering accuracy and cell-type identification as measured by ARI on benchmark datasets.
• Differential expression analysis: Enhances the detection of differentially expressed genes by reducing dropout-related artifacts.
• Method benchmarking: Serves as a benchmark for imputation methods using simulation data and public datasets (Kolod, Pollen, Usoskin).

Methodology:

Computes correlation and negative distance metrics between cells to build a list of similar cells, then imputes dropout values by borrowing gene expression from those similar cells; performance was evaluated using the adjusted rand index (ARI) on simulation and public scRNA-seq datasets.