Related Experiment Video
Updated: Jul 9, 2026
09:19
Spatial Profiling of Protein and RNA Expression in Tissue: An Approach to Fine-Tune Virtual Microdissection
Published on: July 6, 2022
Riemannian metric learning for alignment of spatial multiomics
Peter Halmos1, Yufan Xia1, Benjamin J Raphael1
1Department of Computer Science, Princeton University, Princeton, NJ 08544, United States.
Bioinformatics (Oxford, England)
|July 7, 2026
Summary
Manifold Gromov-Wasserstein (MGW) aligns diverse spatial multiomics data. This novel framework integrates different molecular layers, enabling accurate tissue structure reconstruction and biological discovery across various scales.
Area of Science:
- Computational Biology
- Bioinformatics
- Data Science
Background:
- Emerging spatial technologies enable multi-modal measurements (transcriptome, epigenome, proteome, etc.) from thousands of cells within a tissue.
- Integrating spatial data from heterogeneous feature spaces is challenging, as most assays profile only one modality per tissue slice.
- Existing multi-modal integration techniques often struggle with spatial alignment across diverse data types, particularly when incorporating both spatial and feature information.
Purpose of the Study:
- To develop a novel computational framework for aligning spatial multiomics data from arbitrary modalities.
- To address the challenge of integrating heterogeneous feature spaces in spatial biology.
- To enable robust reconstruction of tissue structures and biological insights from multi-modal spatial datasets.
Main Methods:
- Introduction of Manifold Gromov-Wasserstein (MGW), a metric-learning framework leveraging the product structure of spatial multiomics.
- Inference of modality-specific Riemannian pull-back metrics using neural fields.
- Alignment of Riemannian distances via Gromov-Wasserstein optimal transport, providing a hyperparameter-free cost across modalities with a shared spatial base.
Main Results:
- MGW demonstrates theoretical invariances, including orthogonal transformations and global feature scalings.
- Successful application of MGW to diverse alignment tasks: mouse embryo spatiotemporal transcriptomics (Stereo-Seq), colorectal cancer spatial transcriptomics (Xenium, Visium), and human striatum/kidney cancer spatial metabolomics-transcriptomics.
- MGW recovers biologically meaningful correspondences and spatially coherent tissue structures, outperforming existing optimal transport (OT) and non-OT multi-modal baselines.
Conclusions:
- MGW offers a powerful and versatile solution for spatial multiomics data integration.
- The framework facilitates the alignment of diverse spatial datasets, leading to improved biological interpretation.
- MGW advances the field of spatial multiomics by enabling more comprehensive analysis of tissue architecture and cellular interactions.
