GE-Impute: graph embedding-based imputation for single-cell RNA-seq data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution gene expression data.
• High dropout rates in scRNA-seq introduce artificial zeros, compromising data reliability.
• Existing imputation methods struggle to fully address scRNA-seq data sparsity.

Purpose of the Study:

• To introduce GE-Impute, a novel graph embedding-based neural network for imputing dropout zeros in scRNA-seq data.
• To enhance the accuracy of single-cell expression profiles by mitigating sparsity.
• To improve the biological interpretability of scRNA-seq datasets.

Main Methods:

• GE-Impute utilizes a graph embedding neural network to learn cell representations.
• It reconstructs cell-cell similarity networks for improved neighbor allocation.
• The method is evaluated on both droplet- and plate-based scRNA-seq data.

Main Results:

• GE-Impute demonstrates superior performance in recovering dropout zeros compared to existing methods.
• The approach significantly improves the identification of differentially expressed genes.
• GE-Impute enhances unsupervised clustering, marker gene identification, and trajectory analysis.

Conclusions:

• GE-Impute effectively addresses sparsity in scRNA-seq data, improving data quality.
• The method facilitates more accurate biological interpretation, including cell type assignment and differentiation trajectory reconstruction.
• GE-Impute offers a valuable tool for researchers utilizing scRNA-seq data.