Improved downstream functional analysis of single-cell RNA-sequence data using DGAN.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables high-resolution gene expression analysis.
• Data sparsity and missing values (dropouts) are significant challenges in scRNA-seq data.
• These missing values impede accurate downstream functional analysis.

Purpose of the Study:

• To develop a robust and efficient imputation method for scRNA-seq data.
• To address the issue of missing values and improve downstream analysis outcomes.
• To introduce the deep generative autoencoder network (DGAN) for scRNA-seq data imputation.

Main Methods:

• Designed a deep generative autoencoder network (DGAN) framework.
• DGAN utilizes a variational autoencoder architecture.
• Incorporated count distribution, Gaussian modeling for sparsity, and cell dependency analysis to detect outliers.

Main Results:

• DGAN effectively imputes missing values in sparse gene expression matrices.
• The method demonstrated superior performance compared to baseline methods on five public scRNA-seq datasets.
• Improved outcomes in cell data visualization, clustering, classification, and differential expression analysis were observed.

Conclusions:

• DGAN is a robust and efficient tool for imputing scRNA-seq data dropouts.
• The framework enhances the reliability of downstream functional analyses.
• DGAN offers a valuable solution for a critical challenge in single-cell genomics research.