SCONE: a subset-contrastive method for multi-omics network embedding

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Network-based omics analyses are memory-intensive, limiting their application to smaller datasets.
• Existing multi-omics methods often use similarity networks lacking discrete topologies, reducing graph-based method effectiveness.
• Single-cell omics data analysis requires scalable and effective integration methods.

Purpose of the Study:

• To develop a scalable network-based method for multi-omics data integration.
• To improve cell type clustering in single-cell data using multi-omics integration.
• To address the limitations of existing memory-intensive and topology-restricted omics analysis techniques.

Main Methods:

• Subset-Contrastive multi-Omics Network Embedding (SCONE) utilizes contrastive learning on large datasets.
• A scalable subgraph contrastive approach is employed to handle high-dimensional omics data.
• The method leverages pairwise similarity inherent in omics data for integration.

Main Results:

• SCONE demonstrates synergistic omics integration for cell type clustering in single-cell data.
• The method achieves scalable and effective analysis of large omics datasets.
• SCONE performs comparably to state-of-the-art methods in bulk multi-omics integration, even with limited data views.

Conclusions:

• SCONE provides a scalable and effective approach for multi-omics integration.
• The subset contrastive learning strategy is promising for analyzing large-scale omics data.
• Further research into subset contrastive methods for omics data analysis is warranted.