nsDCC: dual-level contrastive clustering with nonuniform sampling for scRNA-seq data analysis.

Area of Science:

• Computational Biology
• Bioinformatics
• Genomics

Background:

• Single-cell RNA sequencing (scRNA-seq) analysis requires effective dimensionality reduction and clustering.
• Current methods often treat these tasks independently, limiting their synergistic potential.
• Deep clustering methods offer joint optimization but struggle with pre-defined cluster centers.

Purpose of the Study:

• To develop a novel method for joint dimensionality reduction and clustering in scRNA-seq data analysis.
• To leverage contrastive learning to overcome limitations of existing deep clustering approaches.
• To improve the analysis of imbalanced scRNA-seq datasets.

Main Methods:

• Proposed a dual-level contrastive clustering method with nonuniform sampling (nsDCC).
• Integrated instance-level and cluster-level contrastive learning with multi-positive learning and unit matrix constraints.
• Incorporated an attention mechanism for inter-cellular information capture and a sparsest density weight assignment for imbalanced data handling.

Main Results:

• nsDCC outperformed six state-of-the-art methods on real and simulated scRNA-seq data.
• Demonstrated superior performance in both dimensionality reduction and clustering, particularly for imbalanced datasets.
• Showed robustness to 'dropout events' and confirmed biological meaningfulness through differential gene expression analysis.

Conclusions:

• nsDCC offers a powerful and integrated approach for scRNA-seq data analysis.
• The method effectively addresses challenges posed by imbalanced data and technical noise.
• Provides a new framework for understanding complex single-cell transcriptomic data.