TISSUE: uncertainty-calibrated prediction of single-cell spatial transcriptomics improves downstream analyses.

Area of Science:

• Single-cell biology
• Computational biology
• Genomics

Background:

• Whole-transcriptome spatial profiling at single-cell resolution is technically challenging.
• Existing spatial gene expression prediction methods lack reliable uncertainty estimation, impacting downstream analysis quality.

Purpose of the Study:

• To develop a general framework, Transcript Imputation with Spatial Single-cell Uncertainty Estimation (TISSUE), for estimating uncertainty in spatial gene expression predictions.
• To provide uncertainty-aware methods for downstream inference tasks using predicted spatial transcriptomics data.

Main Methods:

• Leveraging conformal inference to generate well-calibrated prediction intervals for gene expression values.
• Benchmarking TISSUE across 11 diverse spatial transcriptomics datasets.
• Applying TISSUE to MERFISH data from the adult mouse subventricular zone.

Main Results:

• TISSUE provides calibrated uncertainty estimates for spatial gene expression predictions.
• It significantly reduces the false discovery rate in differential gene expression analysis.
• Improved clustering, visualization, and supervised learning performance using predicted spatial transcriptomics.
• Identification of neural stem cell subtypes and development of subtype-specific classifiers in the mouse subventricular zone.

Conclusions:

• TISSUE offers a robust solution for uncertainty quantification in spatial gene expression prediction.
• This framework enhances the reliability and interpretability of spatial transcriptomics analyses.
• TISSUE facilitates deeper biological insights, including cell subtype discovery and regional classification.