scSCCNIA: similarity matrix based contrastive clustering with neighbor information aggregation for single-cell RNA...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals cell heterogeneity and gene expression.
• Cell clustering is vital for scRNA-seq data analysis but challenged by high dimensionality and data dropouts.

Purpose of the Study:

• To propose a novel contrastive clustering framework, scSCCNIA, for accurate cell type identification in scRNA-seq data.
• To enhance the accuracy of cell clustering and marker gene identification from complex scRNA-seq datasets.

Main Methods:

• scSCCNIA utilizes a Laplacian filter for neighbor information aggregation.
• Employs Siamese encoders with un-shared parameters for data augmentation and graph view construction.
• Learns latent low-dimensional embeddings through similarity-matrix-based contrastive learning.

Main Results:

• scSCCNIA demonstrates superior performance in cell clustering and marker gene identification across diverse scRNA-seq datasets.
• The framework effectively reveals cell type heterogeneity and functional specificity.
• Enrichment analyses (GO, KEGG) validate the biological insights derived from scSCCNIA.

Conclusions:

• scSCCNIA is an effective algorithm for learning latent features from scRNA-seq data.
• It significantly enhances cell type identification accuracy.
• Facilitates more robust downstream analyses of scRNA-seq data.