scZGA: a novel model based on ZINB distribution and graph attention for scRNA-seq data clustering

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for cell type identification.
• Clustering is a common scRNA-seq analysis technique.
• High dropout rates and complex cell relationships pose significant challenges in scRNA-seq data analysis.

Purpose of the Study:

• To develop a novel computational model for improved scRNA-seq data clustering.
• To address the limitations of existing methods in handling scRNA-seq data characteristics.

Main Methods:

• Proposed a novel model, scZGA, integrating zero-inflated negative binomial (ZINB) distribution and graph attention networks.
• Utilized a ZINB model for global probabilistic structure.
• Employed a graph autoencoder with residual connections for learning neighbor relationships and preserving topological information.
• Implemented a self-optimizing embedding algorithm for deep clustering.

Main Results:

• The scZGA model demonstrated superior performance in scRNA-seq data clustering.
• Achieved higher scores across six diverse scRNA-seq datasets.
• Evaluation metrics including normalized mutual information and adjusted rand index confirmed the model's effectiveness.

Conclusions:

• scZGA offers a robust solution for scRNA-seq data clustering.
• The model effectively overcomes challenges associated with dropout rates and intercellular relationships.
• Results indicate scZGA's potential for advancing cell type identification in single-cell genomics.