ScCLC: A Flexible Contrastive Learning Framework for Single-cell Multi-omics Data Clustering

Area of Science:

• Single-cell biology
• Computational biology
• Bioinformatics

Background:

• Single-cell joint profiling technologies allow simultaneous measurement of multiple molecular modalities.
• Integrating heterogeneous, high-dimensional multi-omics data for cell clustering is challenging.
• Accurate cell clustering is crucial for characterizing cellular heterogeneity.

Purpose of the Study:

• To propose scCLC, a topology-aware contrastive learning framework for single-cell multi-omics data clustering.
• To improve the integration and clustering of multi-modal single-cell data.
• To develop a flexible and scalable method for multi-omics data analysis.

Main Methods:

• scCLC utilizes contrastive learning for cell representation learning.
• A multi-view data augmentation strategy addresses modality-specific characteristics.
• Exploits cell-cell topological structures for self-supervised training and discriminative representations.

Main Results:

• scCLC demonstrates effectiveness in clustering single-cell multi-omics data across multiple datasets.
• Visualization analyses show scCLC can distinguish rare cell populations in imbalanced datasets.
• Case studies confirm scCLC's flexibility and scalability for integrating additional modalities.

Conclusions:

• scCLC provides an effective approach for clustering single-cell multi-omics data.
• The framework accurately identifies cellular heterogeneity and distinguishes rare cell types.
• scCLC is a versatile tool for multi-omics data analysis, adaptable to more modalities.