SAE-Impute: imputation for single-cell data via subspace regression and auto-encoders

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) is vital for studying cellular heterogeneity.
• Dropout events in scRNA-seq data present significant challenges for analysis.
• Existing imputation methods often neglect inter-sample correlations.

Purpose of the Study:

• To introduce SAE-Impute, a novel computational method for imputing scRNA-seq data.
• To enhance the accuracy and reliability of dropout data imputation.
• To improve downstream analysis of scRNA-seq data.

Main Methods:

• SAE-Impute combines subspace regression and autoencoders.
• Subspace regression assesses sample correlations.
• Autoencoders interpolate dropout values using predicted data.

Main Results:

• SAE-Impute reduces false negative signals in scRNA-seq data.
• The method improves the retrieval of dropout values and correlations.
• Downstream analyses like differential gene expression and cell clustering were enhanced.

Conclusions:

• SAE-Impute effectively reduces dropouts in single-cell datasets.
• The imputation method improves the functional interpretability of scRNA-seq data.