HiSTaR: identifying spatial domains with hierarchical spatial transcriptomics variational autoencoder

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) enables transcriptome-wide data acquisition with spatial context.
• Understanding tissue microenvironments and spatial domains is crucial in biological research.
• Deep learning methods are effective for analyzing complex ST data.

Purpose of the Study:

• To introduce HiSTaR, a hierarchical variational autoencoder for ST data.
• To leverage multi-level latent features for enhanced spatial transcriptomics analysis.
• To improve spatial domain identification and batch effect correction in ST data.

Main Methods:

• Developed Hierarchical Spatial Transcriptomics variational autoencoder (HiSTaR).
• Employed multiple HiSTaR blocks to capture multi-level latent features from spatial spots.
• Utilized latent features for downstream analyses like spatial domain identification and batch correction.

Main Results:

• HiSTaR demonstrated superior performance in spatial domain identification across diverse ST datasets.
• The method successfully integrated multiple tissue slices, correcting batch effects without external tools.
• HiSTaR supports trajectory and differential gene expression analyses, validating its effectiveness.

Conclusions:

• HiSTaR provides an effective computational framework for spatial transcriptomics research.
• The hierarchical feature capture improves spatial domain identification and understanding of tissue heterogeneity.
• HiSTaR has the potential to advance the study of spatially resolved gene expression patterns.