CDSImpute: An ensemble similarity imputation method for single-cell RNA sequence dropouts.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for understanding cellular heterogeneity and transcriptomics.
• Technical limitations in scRNA-seq lead to dropout events (excessive zero counts), hindering accurate gene expression analysis.
• Existing methods struggle to reliably distinguish technical dropouts from true biological zeros.

Purpose of the Study:

• To develop a novel method for imputing dropout events in scRNA-seq data.
• To accurately recover true gene expression by distinguishing technical zeros from biological zeros.
• To improve downstream analyses such as differential expression and cell-type identification.

Main Methods:

• Developed CDSImpute (Correlation Distance Similarity Imputation), a method to identify and correct scRNA-seq dropouts.
• Utilizes correlation and negative distance between cells to identify similar cells.
• Borrows gene expression from identified similar cells to impute dropout events.

Main Results:

• CDSImpute successfully identifies technical dropouts and recovers true gene expression.
• Demonstrated improved performance on simulation data and publicly available scRNA-seq datasets.
• Achieved high clustering accuracy (Adjusted Rand Index: 1.00, 0.79, 0.34) on Kolod, Pollen, and Usoskin datasets.
• Outperformed existing methods in precise cell-type identification and differential gene expression detection.

Conclusions:

• CDSImpute is an effective method for imputing dropout events in scRNA-seq expression matrices.
• The method enhances the accuracy of scRNA-seq data analysis.
• CDSImpute is implemented in R and publicly available.