Enhancing Single-Cell RNA-Seq Data Completeness With a Graph Learning Framework.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data.
• Dropout events, characterized by excessive zero counts, are a major challenge in scRNA-seq data.
• These dropouts can obscure true biological signals and hinder downstream analyses.

Purpose of the Study:

• To develop an advanced imputation technique for addressing dropout events in scRNA-seq data.
• To improve the accuracy of gene expression matrices generated from scRNA-seq.
• To enhance the performance of various downstream analyses, including cell clustering and rare cell identification.

Main Methods:

• Developed VAImpute, a variational graph autoencoder-based imputation method.
• Constructed a large network/graph from scRNA-seq data, leveraging copula correlation among cells and genes.
• Utilized the trained model to predict dropout events and impute missing expression values.

Main Results:

• VAImpute demonstrated significant improvements in detecting dropout events compared to existing methods.
• The imputation method achieved superior performance in cell clustering and identifying rare cell populations.
• Enhanced accuracy in differential gene expression analysis was observed using VAImpute.

Conclusions:

• VAImpute offers a robust solution for handling dropout events in scRNA-seq data.
• The method improves the reliability and interpretability of scRNA-seq datasets.
• VAImpute facilitates more accurate biological insights from single-cell gene expression studies.