36 times higher than the original resolution, teams from Beihang University and Tsinghua University used AI to characterize tissues at high resolution on a multi-space omics platform, published in the Nature sub-journal

Spatial omics data enhancement framework soScope

1. Introduction

Spatial omics has expanded the scope of molecular class analysis, but many techniques are limited by spatial resolution. Existing computational methods are mainly targeted at transcriptomic data and lack adaptability to emerging spatial omics technologies.

2. soScope framework

Researchers from Beihang University and Tsinghua University proposed soScope, a unified generation framework designed to improve the quality and resolution of spatial omics data.

3. Technical Principle

soScope summarizes multimodal tissue information from omics, spatial relationships and images. Output omics spectra with enhanced resolution through joint inference of distribution priors and omics-specific modeling.

4. Performance evaluation

The evaluation results of soScope on Visium, Xenium, spatial-CUT&Tag, slide-DNA/RNA-seq and other platforms show that:

• improves the performance of intestinal and kidney structure identification
• Revealed the fine structure of the embryonic heart
• Corrected for sample and technical bias

5. Extended applications

soScope has been extended to spatial-CITE-seq and spatial ATAC-RNA-seq, leveraging cross-omics references for multi- Omic enhancement.

6. Conclusion

soScope provides a versatile tool that improves the utilization of spatial omics technology and resources.

7. Reference

This research was published in "Nature Communications" on August 2, 2024 under the title "Tissue characterization at an enhanced resolution across spatial omics platforms with deep generative model".

Tissue spatial omics technology

Tissues are composed of cells with different molecular states and spatial organizations. Spatial omics technology has made significant progress in recent years, allowing spatial analysis of various molecular classes while maintaining spatial context.

Challenges and Solutions

Despite early success, spatial omics technology still faces two major challenges:

1. Frozen or formalin-fixed tissue may affect the molecular state and reduce sequencing accuracy.
2. Most techniques have limited spatial resolution, making it difficult to reveal subtle heterogeneities in tissue structure.

Computing technology can improve the resolution of spatial omics data, but most current methods only target a single tissue modality, making it difficult to fully utilize multi-modal information.

soScope: Improving spatial resolution and data quality

Research teams from Beihang University and Tsinghua University introduce Spatiomic Scope (soScope), a fully generative framework that simulates point-level data from different spatial omics technologies The profile generation process aims to improve their spatial resolution and data quality.

soScope treats each point as a collection of "sub-points" with enhanced spatial resolution, whose omics characteristics are related to spatial location and morphological patterns. SoScope then uses a multimodal deep learning framework to integrate spot omics profiles, spatial relationships, and high-resolution morphology images and jointly infer omics profiles at sub-spot resolution. By selecting omics-specific distributions, soScope can accurately model and reduce variation in different spatial omics data.

Illustration: Overview of soScope and its applications. (Source: Paper)

soScope Features:

1. Unified tool, combined with multi-modal tissue maps
2. Enhanced omics maps (different molecular classes)
3. Improve spatial resolution
4. Reduce unnecessary changes
5. Characterization Complex tissue structures (undetectable at native resolution)

   

   Illustration: Evaluation of soScope on spatial transcriptomic datasets from multiple tissues and platforms. (Source: Paper)

The team extensively evaluated soScope’s effectiveness and generalizability for multiple molecular types analyzed by multiple spatial techniques, including Visium, Xenium, spatial-CUT&Tag, slide-DNA-seq, slide- RNA-seq, spatial-CITE-seq and spatial ATAC-RNA-seq.

In healthy and diseased tissues, soScope improves tissue domain identification, increases the differentiability of known markers, and corrects for data and technical biases. The method is able to reveal finer tissue structures up to 36 times greater than the original resolution. It can efficiently adapt spatial multi-omics data to simultaneously enhance multi-omics profiles.

Researchers note that there are several imaging-based spatial omics technologies, such as seqFISH, STARmap and MERFISH, which can directly achieve spatial analysis at single-cell resolution, but at the expense of lower omics throughput and smaller tissue areas . Although soScope provides enhanced profiles for prespecified subspot or cellular locations, it may not achieve subcellular resolution.

Illustration: Applying multi-omics soScope to a spatial protein + transcript dataset from human skin tissue using spatial CITE-seq. (Source: paper)

Further improve resolution:

• Modify soScope to include paired single-cell omics data from the same tissue, providing higher resolution information for sub-point inference.
• Integrate H&E images as input, which can be easily annotated by human experts in certain clinical studies.
• Modify soScope to integrate human labels and guide posterior inference in a semi-supervised manner, improving latent representation and profile learning.

Reduce computational costs:

For larger data sets containing multiple contiguous slices from the same organ, soScope can:

• Train on partial data.
• Apply to remaining tissue sections.

Potential:

With the continuous expansion of spatial omics data resources and the emergence of new spatial technologies, researchers believe that soScope has the following potential:

• A versatile tool.
• Make full use of spatial omics data.
• Enhance scientists’ understanding of complex tissue structures and biological processes.

Paper link:
https://www.nature.com/articles/s41467-024-50837-5

The above is the detailed content of 36 times higher than the original resolution, teams from Beihang University and Tsinghua University used AI to characterize tissues at high resolution on a multi-space omics platform, published in the Nature sub-journal. For more information, please follow other related articles on the PHP Chinese website!