Graph-Embedded Deep Generative Clustering for Single-Cell Multi-Omics Data Integration.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell multi-omics technologies generate vast datasets for biological discovery.
• Integrating diverse omics data at the single-cell level is challenging due to data heterogeneity.
• Current methods often fail to leverage cell graph structures effectively.

Purpose of the Study:

• To develop a novel model for effective single-cell multi-omics data integration.
• To address the challenge of high heterogeneity across different omics types.
• To improve the performance and practical utility of multi-omics data analysis.

Main Methods:

• Proposed a Graph-embedded Deep Generative Clustering (GeDGC) model.
• Simultaneously learned shared latent representations and cluster factors using Gaussian mixture models.
• Incorporated graph embedding constraints on latent representations and cluster assignments to preserve local data structure.

Main Results:

• GeDGC effectively captures complex correlations across multiple omics.
• The model generates informative shared latent embeddings for downstream tasks.
• Experimental results on ten datasets demonstrated GeDGC's superiority over seventeen competing methods.

Conclusions:

• GeDGC offers a superior approach for single-cell multi-omics data integration.
• The model's ability to preserve intrinsic cell structure enhances data analysis.
• This method advances the utility of multi-omics data in biological discovery and medical research.