Imputing single-cell RNA-seq data by combining graph convolution and autoencoder neural networks.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) offers high-resolution transcriptome profiling.
• Dropout events, caused by low mRNA capture, are a significant challenge in scRNA-seq data analysis.
• Effective imputation methods are crucial for accurate downstream analyses of scRNA-seq data.

Purpose of the Study:

• To develop a robust imputation method for scRNA-seq data.
• To address the issue of missing data (dropout events) in scRNA-seq experiments.
• To leverage graph convolution networks for improved imputation accuracy.

Main Methods:

• Development of GraphSCI, an imputation method utilizing graph convolution networks.
• Application of GraphSCI to both simulated and real scRNA-seq datasets.
• Evaluation of GraphSCI against existing state-of-the-art imputation techniques.

Main Results:

• GraphSCI demonstrates superior performance in imputing dropout events compared to other methods.
• The method accurately infers gene-to-gene relationships from scRNA-seq data.
• Inferred gene relationships dynamically assist the imputation process during training.

Conclusions:

• GraphSCI is a highly effective tool for imputing missing data in scRNA-seq.
• The ability to infer gene relationships is a key advancement over existing imputation algorithms.
• GraphSCI enhances the reliability and utility of scRNA-seq data for biological discovery.