stAI: a deep learning-based model for missing gene imputation and cell-type annotation of spatial transcriptomics.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) provides RNA transcript levels within their native spatial context.
• Single-cell spatial transcriptomics (scST) offers high-resolution spatial and expression data but has limited transcript detection.
• Accurate whole-transcriptome characterization and cell-type annotation remain challenges in scST.

Purpose of the Study:

• To introduce stAI, a deep learning model for addressing missing gene imputation and cell-type annotation in scST data.
• To enhance the analytical capabilities of scST by enabling comprehensive transcriptome analysis and precise cell identification.

Main Methods:

• stAI employs a joint embedding strategy integrating scST and reference scRNA-seq data.
• Utilizes two separate encoder-decoder modules for imputation and annotation within a supervised latent space.
• Leverages scRNA-seq data to guide the imputation and annotation processes.

Main Results:

• stAI accurately predicts unmeasured genes, including crucial marker genes, in scST datasets.
• Demonstrates high precision in annotating cell types, even for small cell populations.
• Outperforms existing methods in both imputation and annotation tasks across diverse scST platforms.

Conclusions:

• stAI significantly advances scST data analysis by effectively imputing missing gene expression and improving cell-type annotation.
• The model enhances the utility of scST for comprehensive cellular system characterization.
• stAI offers a robust solution for overcoming key limitations in current scST applications.