SmartImpute is a targeted imputation framework for single-cell transcriptome data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) generates high-dimensional data with missing values, complicating downstream analyses.
• Accurate imputation of missing data is crucial for understanding cellular heterogeneity and biological processes.
• Existing imputation methods may not preserve biological nuances or scale effectively to large datasets.

Purpose of the Study:

• To introduce SmartImpute, a novel imputation framework for scRNA-seq data.
• To enhance biological relevance and computational efficiency in imputation by focusing on marker genes.
• To improve the accuracy of scRNA-seq data analysis, including clustering, cell type annotation, and trajectory inference.

Main Methods:

• Development of SmartImpute, a targeted imputation framework utilizing a modified Generative Adversarial Imputation Network (GAIN).
• Incorporation of a multi-task discriminator to impute missing values while preserving true biological zeros.
• Application to diverse scRNA-seq datasets (head and neck squamous cell carcinoma, bone marrow, lung cancer) and spatial transcriptomics data.

Main Results:

• SmartImpute demonstrated improved performance in clustering, cell type annotation, and trajectory inference across multiple scRNA-seq datasets.
• The framework successfully scaled to handle datasets exceeding one million cells.
• Application to spatial transcriptomics data revealed enhanced spatial gene expression patterns and clustering.

Conclusions:

• SmartImpute provides a robust and efficient solution for imputing missing data in scRNA-seq and spatial transcriptomics.
• The targeted approach focusing on marker genes improves biological interpretability and analytical outcomes.
• SmartImpute facilitates deeper insights into cellular heterogeneity, disease progression, and spatial transcriptomic data organization.