scMDCL: A Deep Collaborative Contrastive Learning Framework for Matched Single-Cell Multiomics Data Clustering.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell multiomics clustering is vital for understanding cellular heterogeneity and disease mechanisms.
• Current methods often fail to fully exploit intercellular relationships and cross-omics feature synergy, limiting clustering performance.

Purpose of the Study:

• To develop a deep collaborative contrastive learning framework (scMDCL) for enhanced matched single-cell multiomics data clustering.
• To improve the integration of multiomics data by leveraging intercell relationships and inter-omics feature interactions.

Main Methods:

• Proposed a deep collaborative contrastive learning framework (scMDCL).
• Incorporated a graph autoencoder and feature enhancement module to extract and augment cell features from different omics.
• Utilized contrastive learning to strengthen interactions among features from the same cell across different omics.
• Employed multiomics deep collaborative clustering modules for the final clustering.

Main Results:

• The scMDCL framework effectively leverages intercell relationships and enhances feature interactions across omics.
• Demonstrated superior performance in integrating multiomics data for clustering tasks.
• Achieved state-of-the-art results on nine publicly available single-cell multiomics datasets.

Conclusions:

• scMDCL offers a powerful approach for single-cell multiomics data clustering.
• The framework significantly improves the analysis of cellular heterogeneity by maximizing the utility of multiomics data.
• Highlights the potential of deep collaborative contrastive learning in advancing multiomics data integration and interpretation.