ScGDCF: Graphical Deep Clustering with Fused Common Information for Single-cell RNA-seq Data.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Unsupervised deep clustering is vital for identifying cell types in single-cell RNA sequencing (scRNA-seq) data.
• Existing methods struggle with fusing feature and topological information and performing well on sparse scRNA-seq data.

Purpose of the Study:

• To develop a novel deep clustering method, scGDCF, for accurate cell type identification in large and sparse scRNA-seq datasets.
• To effectively fuse common information from feature and topological structures for improved clustering performance.

Main Methods:

• Introduced a sparse feature representation method using an adversarial loss function to address data sparsity.
• Designed a mutual information extracting operator to mine and fuse common information from feature and topological data.
• Incorporated a dual adaptive attention mechanism for global and local information integration.

Main Results:

• scGDCF demonstrated superior performance compared to 17 state-of-the-art methods on seven real-world and two simulated scRNA-seq datasets.
• The method accurately segregated cell types even in large and sparse datasets.
• Extended clustering results provided novel insights into cell developmental lineages and preserved inter-cluster distances.

Conclusions:

• scGDCF offers an effective solution for unsupervised deep clustering of scRNA-seq data, particularly for sparse datasets.
• The method's ability to fuse diverse data information and its attention mechanism contribute to its high performance.
• scGDCF facilitates deeper biological insights through visualization and differential expression analysis.