ST2HE: enhancing spatial transcriptomics interpretability via virtual staining for histological annotation

Area of Science:

• Computational pathology
• Bioinformatics
• Genomics

Background:

• High-resolution spatial transcriptomics (HR-ST) provides deep tissue architecture insights.
• Standardized histological annotation frameworks for HR-ST data are currently lacking.

Purpose of the Study:

• To introduce ST2HE, a cross-platform generative framework for synthesizing virtual hematoxylin and eosin (H&E) images from HR-ST data.
• To enable histologically informative image generation across diverse tissue types and HR-ST platforms.

Main Methods:

• ST2HE integrates nuclei morphology and spatial transcript coordinates using a one-step diffusion model.
• Conditional and tissue-independent variants were developed to support known and novel tissue contexts.
• Evaluations were performed on breast cancer, non-small cell lung cancer, and Kaposi's sarcoma datasets.

Main Results:

• ST2HE successfully preserves morphological features and supports downstream annotations of tissue histology.
• The framework enables accurate phenotype classification from synthesized H&E images.
• Ablation studies identified key parameters like context windows and loss functions that enhance image fidelity.

Conclusions:

• ST2HE bridges the gap between molecular and histological data domains.
• The framework offers interpretable and scalable annotation of HR-ST data.
• ST2HE advances computational pathology by integrating spatial transcriptomics with histological visualization.