ScCCL: Single-Cell Data Clustering Based on Self-Supervised Contrastive Learning.

Area of Science:

• Computational Biology
• Genomics
• Data Science

Background:

• Single-cell RNA sequencing (scRNA-seq) enables detailed cellular-level analysis of complex biological systems.
• Clustering scRNA-seq data is crucial for understanding tissue heterogeneity and disease, but faces challenges from high dimensionality, large cell numbers, and technical noise.

Purpose of the Study:

• To introduce ScCCL, a novel self-supervised contrastive learning method for improved scRNA-seq data clustering.
• To overcome the limitations of existing clustering methods in handling complex single-cell data.

Main Methods:

• ScCCL employs data augmentation by masking gene expression and adding Gaussian noise.
• It utilizes a momentum encoder for feature extraction and applies contrastive learning at both instance and cluster levels.
• The method generates a representation model for high-order single-cell embeddings.

Main Results:

• ScCCL demonstrated improved clustering performance on multiple public datasets compared to benchmark algorithms.
• Evaluation metrics such as Adjusted Rand Index (ARI) and Normalized Mutual Information (NMI) confirmed the effectiveness of ScCCL.
• The method's robustness was highlighted by its applicability to single-cell multi-omics data.

Conclusions:

• ScCCL offers a powerful and versatile approach for clustering scRNA-seq data, outperforming existing methods.
• Its self-supervised contrastive learning framework effectively handles data complexities and noise.
• The method's adaptability makes it valuable for diverse single-cell analyses, including multi-omics integration.