Contrastively generative self-expression model for single-cell and spatial multimodal data.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Single-cell multi-omics technologies offer deep insights into cellular heterogeneity.
• Integrating diverse omics data modalities presents significant computational challenges due to measurement variability.

Purpose of the Study:

• To develop a robust computational framework for integrating heterogeneous single-cell multimodal data.
• To address the challenge of combining omics data with potentially weak inter-modal relationships.

Main Methods:

• Proposed a contrastive and generative deep self-expression model named single-cell multimodal self-expressive integration (scMSI).
• scMSI learns omics-specific latent representations and self-expression relationships using deep generative models.
• Employs contrastive learning to integrate these relationships into a unified manifold space.

Main Results:

• scMSI effectively integrates heterogeneous multimodal data into a unified representation.
• Demonstrated scMSI's utility in various analytical tasks including integration, denoising, batch correction, and spatial domain detection.
• Validated scMSI's high effectiveness and robustness across diverse single-cell and spatial multimodal datasets.

Conclusions:

• scMSI provides a powerful and flexible paradigm for multimodal single-cell data integration.
• The model successfully achieves representation learning and data integration within a single framework.
• scMSI offers a cohesive solution for complex single-cell data analysis challenges.