Single-cell RNA-seq data analysis using graph autoencoders and graph attention networks.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• High-throughput sequencing has led to a surge in single-cell RNA sequencing (scRNA-seq) data.
• scRNA-seq data is characterized by high dimensionality, dropout noise, and sparsity.
• Existing statistical and machine learning methods struggle with scRNA-seq data efficiency and accuracy, while deep learning methods cannot directly process non-Euclidean spatial data.

Purpose of the Study:

• To develop an advanced computational model for scRNA-seq data analysis.
• To improve gene imputation and cell clustering accuracy for scRNA-seq data.
• To address the limitations of existing methods in handling complex scRNA-seq data structures.

Main Methods:

• Developed scGAEGAT, a multi-modal model integrating graph autoencoders and graph attention networks.
• Utilized graph neural networks for scRNA-seq data analysis.
• Employed cosine similarity, median L1 distance, and root-mean-squared error for gene imputation performance evaluation.
• Used adjusted mutual information, normalized mutual information, completeness score, and Silhouette coefficient score for cell clustering performance evaluation.

Main Results:

• scGAEGAT demonstrated promising performance in gene imputation tasks.
• The model achieved high accuracy in cell clustering predictions.
• Experimental results were validated on four scRNA-seq datasets with gold-standard cell labels.
• scGAEGAT outperformed existing methods in key performance metrics.

Conclusions:

• scGAEGAT is an effective graph neural network-based model for scRNA-seq data analysis.
• The model shows significant potential for advancing gene imputation and cell clustering in single-cell genomics.
• This approach offers a robust solution for handling the complexities of scRNA-seq data.