scSFCL:Deep clustering of scRNA-seq data with subspace feature confidence learning

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis for cell type diversity.
• High dimensionality, sparsity, and noise in scRNA-seq data challenge traditional clustering methods.
• Existing methods struggle to fully utilize discriminative attribute information and accurately capture cell diversity.

Purpose of the Study:

• To propose a novel deep clustering method, scSFCL, for scRNA-seq data.
• To address the limitations of traditional clustering in handling complex single-cell data.
• To improve the accuracy and effectiveness of cell type identification from scRNA-seq data.

Main Methods:

• Developed scSFCL based on subspace feature confidence learning.
• Utilized kernel density to divide subspaces and filter discriminative feature subsets.
• Employed a graph convolutional network (GCN) with weighting for feature confidence learning.
• Integrated GCN and a denoising variational autoencoder based on zero-inflated negative binomials (DVAE-ZINB) for mutually supervised clustering.

Main Results:

• scSFCL demonstrated significantly improved clustering performance on multiple scRNA-seq datasets.
• The method effectively filters discriminative feature subsets and learns their confidence.
• Complementary fusion of structural and idiosyncratic information enhanced clustering accuracy.

Conclusions:

• scSFCL provides an effective solution for deep clustering of scRNA-seq data.
• The proposed method overcomes limitations of traditional approaches in capturing cell type diversity.
• Enhanced feature learning and information fusion contribute to superior clustering performance.