Related Experiment Video
Updated: Aug 6, 2026
09:06
Spatially Resolved, Integrated Single-Cell Multiomic Profiling of the Transcriptome and Epigenomic Targets in Frozen Tissue Sections
Published on: June 12, 2026
PRISM: Prior-enhanced Inference for Spatial Transcriptomic Cell Type Mapping
Yiheng Xu1,2,3, Xuehao Wang3, Shuqi Liu1,4,2
1Department of Psychiatry of the Second Affiliated Hospital, Zhejiang University School of Medicine, Hangzhou, 310009, China.
Bioinformatics (Oxford, England)
|July 23, 2026
Summary
PRISM enhances spatial transcriptomics cell type annotation by integrating biological priors and self-training. This novel framework improves accuracy and robustness across diverse datasets and platforms.
Area of Science:
- Computational Biology
- Genomics
- Bioinformatics
Background:
- Cell type annotation in spatial transcriptomics (ST) is crucial for understanding tissue organization and biological processes.
- Existing methods often rely on single-cell RNA-seq (scRNA) data but struggle with domain gaps and spatial dependencies.
- Marker gene selection is frequently treated as a separate step, leading to instability and limited interpretability.
Purpose of the Study:
- To develop a robust framework for spatial transcriptomics cell type annotation that addresses limitations of current methods.
- To improve the accuracy and biological interpretability of cell type mapping in ST data.
- To create a method that is resilient to domain shifts and platform-specific noise.
Main Methods:
- PRISM (Prior-enhanced Inference for Spatial Transcriptomic Cell Type Mapping) is a three-stage framework.
- It integrates biological prior construction for marker gene extraction and a prior-enhanced self-training strategy.
- The framework refines predictions by encoding spatial information and optimizing under biological constraints.
Main Results:
- PRISM demonstrates strong performance on eleven ST datasets across six platforms and two species.
- It achieves high Accuracy and Macro-F1 scores on labeled benchmarks for both brain and non-brain tissues.
- In label-free settings, PRISM shows superior robustness to domain shift and platform heterogeneity, achieving the best composite rank.
Conclusions:
- PRISM offers a significant advancement in spatial transcriptomics cell type annotation.
- The framework provides stable and interpretable predictions by leveraging biological priors and spatial information.
- PRISM effectively bridges the domain gap between scRNA and ST data, enabling reliable cell type mapping across diverse experimental conditions.
