Single-cell RNA-seq data analysis based on directed graph neural network.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) generates large datasets but faces challenges like data sparsity and complex gene expression patterns.
• Traditional statistical and machine learning methods struggle with scRNA-seq data efficiency and accuracy.
• Existing deep learning methods cannot directly process non-Euclidean spatial data from cell diagrams.

Purpose of the Study:

• To develop a novel deep learning framework for enhanced scRNA-seq data analysis.
• To address limitations of existing methods in handling sparsity and spatial data.
• To improve gene imputation and cell clustering performance in scRNA-seq data.

Main Methods:

• Development of scDGAE, a directed graph neural network incorporating graph autoencoders and graph attention networks.
• Utilizing directed graph neural networks to preserve graph connectivity and expand receptive fields.
• Performance evaluation using metrics like cosine similarity, L1 distance, RMSE for gene imputation, and AMI, NMI, completeness, Silhouette scores for cell clustering.

Main Results:

• scDGAE demonstrated promising performance in gene imputation across four scRNA-seq datasets.
• The model achieved high accuracy in cell clustering prediction on datasets with gold-standard labels.
• Comparative analysis showed scDGAE outperforming existing methods in key performance metrics.

Conclusions:

• scDGAE offers a robust and effective framework for scRNA-seq data analysis.
• The model successfully addresses challenges of sparsity and improves imputation and clustering.
• scDGAE provides a versatile tool applicable to a broad range of scRNA-seq analyses.