Cross-modality representation and multi-sample integration of spatially resolved omics data.

PRESENT, a contrastive learning framework for spatial multi-omics, enables accurate identification of spatial domains and elucidation of underlying regulatory mechanisms across diverse species and technologies.

• Why it matters: Spatially resolved omics integration is hindered by heterogeneity across modalities and samples, limiting reliable cross-modality and cross-sample interpretation of tissue microenvironments.
• What they did: PRESENT uses omics-specific encoders with graph attention networks and Bayesian neural networks plus distribution-aware decoders, and an inter-omics alignment module; it extends to multi-sample integration with two-stage training using inter-batch alignment, intra-batch preserving, batch-adversarial learning, and cyclic graph refinement.
• The result: Experiments on tissue samples across different anatomical regions and developmental stages show PRESENT characterizes hierarchical tissue structures from a spatiotemporal perspective.

AI summary of the abstract · may contain errors · not medical advice · read the paper.