Decoupled GNNs based on multi-view contrastive learning for scRNA-seq data clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Clustering single-cell RNA sequencing (scRNA-seq) data is crucial for understanding cellular heterogeneity.
• Existing methods face challenges with high dimensionality and complexity, particularly graph neural networks (GNNs) which suffer from computational complexity due to exponential dependency growth.
• Inefficient GNN training hinders the effective analysis of scRNA-seq data.

Purpose of the Study:

• To develop an efficient and accurate clustering method for scRNA-seq data.
• To address the computational complexity and performance limitations of GNNs in scRNA-seq clustering.
• To improve the identification of cell types and states from complex single-cell data.

Main Methods:

• Proposed scDeGNN, a novel approach utilizing decoupled graph neural networks (GNNs) and multi-view contrastive learning for scRNA-seq data clustering.
• Constructed two distinct views using adjacency matrices and employed decoupled GNNs for initial cell feature representation.
• Refined features using a multilayer perceptron and contrastive learning, followed by fusion for the final clustering task.

Main Results:

• scDeGNN demonstrated superior performance compared to state-of-the-art scRNA-seq clustering algorithms.
• Evaluated on nine diverse real-world scRNA-seq datasets across various organisms and tissues.
• Achieved significant improvements across multiple evaluation metrics, highlighting its effectiveness and robustness.

Conclusions:

• scDeGNN offers an effective solution for scRNA-seq data clustering by mitigating GNN computational challenges.
• The multi-view contrastive learning framework enhances feature representation, leading to more accurate cell clustering.
• This method provides a significant advancement for deciphering cellular heterogeneity in single-cell genomics research.