Related Experiment Video
Updated: Aug 5, 2026
10:16
Mining Spatial Transcriptomics Datasets using DeepSpaceDB
Published on: September 5, 2025
DSSMST: A Deterministic State Space Model for Self-Supervised Spatial Domain Identification in Spatial
Jirui Zhang1, Xingyu Liu1, Maoyuan Zhou1
1Academy of Artificial Intelligence, Beijing Institute of Petrochemical Technology, Beijing, 102617, China.
Biochemical Genetics
|July 27, 2026
Summary
We developed DSSMST, a new self-supervised learning framework for spatial transcriptomics data. It accurately identifies functional regions in tissues, advancing spatial domain analysis.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Spatial transcriptomics (ST) enables tissue-level cellular heterogeneity and architecture analysis.
- Accurate identification of functional regions in complex ST data remains challenging.
Purpose of the Study:
- Introduce DSSMST, a novel self-supervised learning framework for ST data analysis.
- Improve spatial domain identification accuracy and robustness.
Main Methods:
- Integrates a Deterministic State Space Model (DSSM) for dynamic modeling of spatial gradients.
- Utilizes graph neural networks (GNNs) and contrastive learning for refined feature embedding.
- Employs a self-supervised contrastive learning mechanism to enhance domain distinction.
Main Results:
- DSSMST demonstrates leading accuracy in spatial domain identification across multiple ST datasets.
- The framework shows competitive robustness and generalization capabilities.
- DSSMST effectively captures spatial gradient variations and continuous dependencies.
Conclusions:
- DSSMST offers a powerful approach for precise spatial domain delineation in ST data.
- The method overcomes limitations of static graph-based approaches.
- DSSMST has strong potential to advance spatial transcriptomics research.
