Deep cross-omics cycle attention model for joint analysis of single-cell multi-omics data

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Joint profiling of single-cell transcriptomics and epigenomics is crucial for understanding cellular heterogeneity and regulatory programs.
• Significant challenges exist in integrating multi-omics data due to differences in sparsity, heterogeneity, and dimensionality.

Purpose of the Study:

• To develop a computational tool for the joint analysis of single-cell multi-omics data.
• To address the limitations hindering integrative analysis of diverse omics datasets.

Main Methods:

• Proposed the deep cross-omics cycle attention (DCCA) model.
• Combined variational autoencoders (VAEs) with attention-transfer mechanisms.
• Leveraged parallel multi-omics data from the same cell to fine-tune networks.

Main Results:

• DCCA demonstrates superior capability in dissecting cellular heterogeneity.
• The model excels at denoising and aggregating multi-omics data.
• DCCA effectively links multi-omics data to infer novel transcriptional regulatory relations.
• The model can generate missing data or omics in a biologically meaningful way.

Conclusions:

• DCCA provides a powerful new approach for analyzing and understanding complex biological processes using single-cell multi-omics data.
• The model facilitates deeper insights into cellular states and regulatory mechanisms.