scBGEDA: deep single-cell clustering analysis via a dual denoising autoencoder with bipartite graph ensemble...

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) is vital for transcriptomic analysis.
• Cell-type clustering is essential for scRNA-seq data analysis.
• Existing deep autoencoders struggle with scRNA-seq latent representations and consensus clustering.

Purpose of the Study:

• To develop scBGEDA, a single-cell deep clustering model.
• To improve latent representation learning and ensemble clustering for scRNA-seq data.
• To accurately identify cell types and characterize transcriptomic profiles.

Main Methods:

• Proposed a single-cell dual denoising autoencoder for compressed low-dimensional space projection.
• Employed synergistic optimization of reconstruction losses for feature representation.
• Designed a bipartite graph ensemble clustering algorithm for cell-cell relationships and latent space exploitation.
• Utilized a graph-based consensus function for robust clustering.

Main Results:

• scBGEDA outperforms state-of-the-art methods across 20 scRNA-seq datasets.
• Demonstrated scalability to large-scale scRNA-seq datasets.
• Successfully identified cell-type specific marker genes and provided functional genomic insights.

Conclusions:

• scBGEDA offers a robust and scalable solution for scRNA-seq data analysis.
• The model enhances cell-type identification and characterization.
• Provides new perspectives for understanding transcriptomic data.