INLAomics for Scalable and Interpretable Spatial Multiomic Data Integration.

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Spatial multi-omics integration is crucial for understanding tissue architecture and biological regulation.
• Current methods often ignore spatial context, limiting interpretability and scalability.
• Existing approaches struggle with dimensionality reduction and autoencoders for multi-modal data.

Purpose of the Study:

• To develop a novel multivariate hierarchical Bayesian framework, INLAomics, for integrating spatial transcriptomics and proteomics data.
• To address limitations of current methods by incorporating histological features and spatial factors.
• To enable accurate modeling of protein abundance within intact tissue architecture.

Main Methods:

• Developed INLAomics, a multivariate hierarchical Bayesian framework.
• Leveraged histological features and latent spatial factors from spatial transcriptomics.
• Modeled protein abundance in tissue sections for integrative analysis.

Main Results:

• INLAomics identified novel spatial gene co-expression programs and gene-protein associations.
• The framework significantly improved spatial protein expression prediction accuracy.
• Demonstrated computational efficiency and biological interpretability across diverse datasets.

Conclusions:

• INLAomics provides a scalable and interpretable solution for spatial multi-omics integration.
• The framework enhances understanding of biological regulation within tissue context.
• Offers a powerful tool for discovering gene-protein relationships and predicting protein expression.