scDCCA: deep contrastive clustering for single-cell RNA-seq data based on auto-encoder network.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables cellular heterogeneity exploration but faces challenges due to high dimensionality, noise, and sparsity.
• Existing cell clustering methods struggle to fully capture intrinsic cellular properties and relationships, impacting downstream analysis performance.
• Accurate cell identity recognition through clustering is crucial for interpreting scRNA-seq data.

Purpose of the Study:

• To propose a novel deep contrastive clustering algorithm, scDCCA, for improved scRNA-seq data analysis.
• To enhance feature extraction and cell relationship modeling for robust cell clustering.
• To achieve end-to-end representation learning and cell clustering.

Main Methods:

• Developed scDCCA, integrating a denoising auto-encoder and a dual contrastive learning module within a deep clustering framework.
• Utilized a denoising Zero-Inflated Negative Binomial model-based auto-encoder for robust low-dimensional feature extraction.
• Incorporated a dual contrastive learning module to capture instance-level and cluster-level cell proximity, enhancing feature discriminability.

Main Results:

• scDCCA demonstrated superior performance across 14 real scRNA-seq datasets compared to eight state-of-the-art methods.
• The algorithm showed improvements in accuracy, generalizability, scalability, and efficiency.
• Cell visualization and biological analyses confirmed enhanced clustering and facilitated downstream interpretation of scRNA-seq data.

Conclusions:

• scDCCA effectively addresses the challenges of high dimensionality, noise, and sparsity in scRNA-seq data.
• The proposed deep contrastive clustering approach significantly improves cell clustering and downstream analysis.
• scDCCA offers a robust and efficient tool for scRNA-seq data interpretation.