SELF-Former: multi-scale gene filtration transformer for single-cell spatial reconstruction.

Area of Science:

• Computational biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) aims to map gene expression to tissue locations.
• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression data but lacks spatial information.
• Challenges include batch effects and precise gene characterization for spatial mapping.

Purpose of the Study:

• To develop a robust computational framework for accurate spatial reconstruction of scRNA-seq data into ST.
• To address challenges in aligning gene expression profiles with spatial origins.
• To mitigate batch effects inherent in scRNA-seq and ST data integration.

Main Methods:

• Developed SELF-Former, a transformer-based framework utilizing multi-scale structures for gene representation learning.
• Incorporated spatial correlation constraints for reconstructing ST data.
• Introduced a novel gene filtration module to select key genes for spatial reconstruction.

Main Results:

• SELF-Former successfully recovered spatial information from ST data.
• The framework effectively mitigated batch effects between scRNA-seq and ST datasets.
• Validated superior performance across four benchmark datasets, demonstrating robust spatial reconstruction capabilities.

Conclusions:

• SELF-Former provides a reliable method for spatial reconstruction, accurately mapping scRNA-seq data to spatial contexts.
• The gene filtration module significantly enhances reconstruction accuracy.
• Represents a significant advancement in integrating gene expression with spatial information in tissues.