I-Impute

I-Impute performs imputation of dropout events in single-cell RNA-seq (scRNA-seq) data to recover gene expression and improve downstream analyses such as cell-type identification and clustering.

Key Features:

• Self-Consistency Principle: Enforces iterative refinement until no further dropouts or errors are detected, producing self-consistent imputed data.
• Continuous Similarities and Dropout Probabilities: Leverages continuous similarities among cells and dropout probability estimates to more accurately estimate missing values and de-noise expression matrices.
• Iterative Refinement: Applies repeated imputation updates to progressively improve the quality of the recovered expression values.

Scientific Applications:

• Cell subtype identification: Enables identification of underlying cell subtypes within real scRNA-seq datasets by recovering missing expression signals.
• Clustering performance improvement: Improves clustering accuracy and normalized mutual information for downstream population identification tasks.
• Analysis of cellular heterogeneity: Facilitates investigation of transcriptomic diversity and cellular heterogeneity in scRNA-seq studies.

Methodology:

Leverages continuous cell–cell similarities and dropout probabilities with iterative refinement to self-consistency, and evaluates imputed data against ground truth using Pearson correlation and other statistical measures in in silico experiments across dropout rates (90.87%, 70.98%, 56.65%), with comparisons to SAVER and scImpute and evaluation on wetlab datasets (mouse bladder cells, embryonic stem cells, aortic leukocyte cells) using adjusted Rand index and normalized mutual information.