Related Experiment Video
Updated: Jul 15, 2026
10:16
Mining Spatial Transcriptomics Datasets using DeepSpaceDB
Published on: September 5, 2025
SPADE: Spatial transcriptomics and pathology alignment using a mixture of data experts for an expressive latent space
Ekaterina Redekop1, Mara Pleasure2, Zichen Wang1
1Biomedical AI Research Lab, University of California, 924 Westwood Blvd, Los Angeles, 90024, CA, United States; Department of Radiological Sciences, University of California, Los Angeles, CA, United States; Department of Bioengineering, University of California, Los Angeles, CA, United States.
Medical Image Analysis
|July 13, 2026
Summary
SPADE, a new foundation model, integrates whole-slide images and spatial transcriptomics data. This approach enhances understanding of molecular heterogeneity in digital pathology, improving disease analysis.
Area of Science:
- Computational pathology
- Genomics
- Artificial intelligence in medicine
Background:
- Digital pathology and self-supervised learning have advanced pathology task models.
- Multimodal approaches exist, but integrating whole-slide images (WSIs) with spatial transcriptomics (ST) remains a gap for molecular heterogeneity analysis.
Purpose of the Study:
- Introduce SPADE, a foundation model for integrating histopathology and ST data.
- Create an ST-informed latent space for unified image representation learning.
- Address the critical gap in combining WSI and ST data for comprehensive analysis.
Main Methods:
- SPADE utilizes a mixture-of-data experts technique.
- Experts are generated through two-stage imaging feature-space clustering with contrastive learning.
- Learns representations from co-registered WSI patches and gene expression profiles.
Main Results:
- SPADE was pre-trained on the HEST-1k dataset.
- Evaluated on 20 downstream tasks, SPADE demonstrated superior few-shot performance.
- Significantly outperformed baseline models by integrating morphological and molecular information.
Conclusions:
- Integrating morphological and molecular data in a unified latent space offers significant benefits.
- SPADE represents a novel foundation model for advancing pathology research.
- The model's performance highlights the potential of combined WSI and ST data analysis.
