VISTA uncovers missing gene expression and spatial-induced information for spatial transcriptomic data analysis.

Area of Science:

• Spatial transcriptomics
• Computational biology
• Genomics

Background:

• Understanding cellular activities in spatial context is vital for deciphering spatially influenced cellular states.
• Single-cell RNA sequencing (scRNA-seq) offers comprehensive gene expression but lacks spatial resolution.
• Subcellular spatial transcriptomics (SST) provides spatial information but covers a limited number of genes.

Purpose of the Study:

• To develop a computational model, VISTA, for predicting unmeasured gene expression in SST data.
• To integrate scRNA-seq and SST data for enhanced spatial transcriptomics analysis.
• To improve the interpretability and utility of SST data by combining molecular coverage with spatial precision.

Main Methods:

• VISTA employs variational inference and geometric deep learning.
• The model integrates scRNA-seq and SST data.
• Uncertainty quantification is incorporated into the deep learning framework.

Main Results:

• VISTA demonstrates superior imputation accuracy, scalability, and efficiency across four diverse datasets.
• Accurate gene expression imputation supports various downstream analyses.
• The model successfully disentangles spatial versus intrinsic expression variations.

Conclusions:

• VISTA effectively bridges the gap between comprehensive gene expression and spatial resolution in transcriptomics data.
• The model enhances the interpretability and utility of SST data for studying tissue organization and cellular microenvironments.
• Accurate imputation by VISTA facilitates advanced analyses like ligand-receptor interactions and spatial RNA velocity inference.