scDSSC: Deep Sparse Subspace Clustering for scRNA-seq Data.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) provides high-resolution transcriptomic data, crucial for understanding cellular heterogeneity.
• Clustering is essential for analyzing scRNA-seq data but faces challenges due to high noise and dimensionality.
• Subspace clustering offers a way to uncover data structures in an unsupervised manner.

Purpose of the Study:

• To propose a novel deep sparse subspace clustering method, scDSSC, tailored for scRNA-seq data.
• To enhance noise reduction and dimensionality reduction capabilities for improved single-cell data analysis.
• To facilitate better interpretability of clustering results and downstream analyses.

Main Methods:

• Developed scDSSC, a deep sparse subspace clustering algorithm.
• Integrated noise reduction and dimensionality reduction within the clustering framework.
• Employed explicit modeling of scRNA-seq data generation for simultaneous feature representation and clustering.

Main Results:

• scDSSC demonstrated significant improvements in clustering performance across various scRNA-seq datasets.
• The method effectively handled datasets ranging from thousands to tens of thousands of cells.
• scDSSC outperformed existing state-of-the-art methods in clustering accuracy and interpretability.

Conclusions:

• scDSSC offers a robust and effective approach for analyzing high-dimensional, noisy scRNA-seq data.
• The method enhances the discovery of cellular heterogeneity through improved clustering.
• scDSSC facilitates more reliable downstream analyses in single-cell research.