I-Impute: a self-consistent method to impute single cell RNA sequencing data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for cell-specific transcriptome studies.
• Gene dropouts are a common challenge in scRNA-seq data, necessitating imputation.
• Current imputation methods often rely on known cell labels, limiting their applicability.

Purpose of the Study:

• To develop a self-consistency based imputation strategy for scRNA-seq data.
• To introduce I-Impute, a method that refines imputation until data consistency is achieved.
• To evaluate I-Impute's performance without requiring prior knowledge of cell subgroups.

Main Methods:

• Proposed a "self-consistency" principle for scRNA-seq data imputation.
• Developed I-Impute, an iterative method optimizing continuous similarities and dropout probabilities.
• Evaluated I-Impute on both in silico and three wet-lab datasets (mouse bladder, embryonic stem, and aortic leukocyte cells).

Main Results:

• I-Impute demonstrated superior Pearson correlations compared to SAVER and scImpute across various dropout rates on in silico data.
• Achieved the highest clustering accuracy on all three wet-lab datasets.
• Showcased effective cell subpopulation discovery without ground-truth labels.

Conclusions:

• The self-consistency strategy, implemented in I-Impute, yields superior imputation results for scRNA-seq data.
• I-Impute outperforms state-of-the-art imputation tools.
• Source code for I-Impute is publicly available for research use.