Interpretable spatial multi-omics data integration and dimensionality reduction with SpaMV

Area of Science:

• Computational biology
• Genomics
• Systems biology

Background:

• Spatial multi-omics technologies offer high-resolution molecular data.
• Current integration methods often project diverse omics into a single latent space, losing unique information.
• This loss limits the full potential of multi-omics analyses.

Purpose of the Study:

• To develop a novel representation learning algorithm for spatial multi-omics data.
• To explicitly capture both shared and omics-specific information.
• To improve the interpretability and comprehensiveness of multi-omics data integration.

Main Methods:

• Developed the Spatial Multi-View (SpaMV) representation learning algorithm.
• SpaMV disentangles shared and omics-specific features from integrated spatial multi-omics data.
• Evaluated performance on simulated and real-world datasets.

Main Results:

• SpaMV achieved superior spatial domain clustering compared to existing methods.
• The algorithm provides more interpretable dimensionality reduction for topic modeling.
• SpaMV effectively identifies omics-specific biomarkers, outperforming current approaches.

Conclusions:

• SpaMV offers a more comprehensive and interpretable approach to spatial multi-omics data integration.
• The method enhances the discovery of omics-specific biomarkers.
• SpaMV unlocks deeper biological insights by preserving unique omics information.