Clustering CITE-seq data with a canonical correlation-based deep learning method.

Area of Science:

• Single-cell biology
• Computational biology
• Genomics

Background:

• Single-cell multiomics sequencing, including CITE-seq, offers comprehensive cellular analysis.
• CITE-seq data presents challenges like noise, high dimensionality, sparsity, and low correlation between RNA and protein data.
• Existing clustering methods struggle with the divergent contributions of RNA and protein data.

Purpose of the Study:

• To develop a robust clustering method for single-cell multiomics data, specifically CITE-seq.
• To effectively integrate gene expression and surface protein data for improved cell state and interaction analysis.
• To address the challenges of noise, sparsity, and divergent data contributions in CITE-seq data.

Main Methods:

• Proposed scCTClust, a novel computational framework for multiomics data clustering.
• Employed omics-specific neural networks to extract cluster information from RNA and protein data.
• Utilized deep canonical correlation analysis for integrating multiomics representations and a decentralized clustering approach with adaptive fusion weights.

Main Results:

• scCTClust demonstrated superior performance in clustering simulated and real CITE-seq datasets.
• The method successfully integrated gene expression and surface protein data, overcoming low correlation issues.
• Adaptive fusion weights effectively balanced the contributions of different omics for clustering.

Conclusions:

• scCTClust provides a powerful and effective approach for clustering single-cell CITE-seq data.
• The method enhances the understanding of cell states and interactions by integrating multiomics information.
• scCTClust shows potential for generalization to other single-cell multiomics datasets, such as transcriptome-epigenome data.