Multiscale confidence quantification for virtual spatial transcriptomics with UTOPIA.

Area of Science:

• Genomics
• Computational Biology
• Bioinformatics

Background:

• Virtual spatial transcriptomics (ST) methods predict gene expression and cell types from histology images.
• These methods extend molecular insights beyond directly measured ST regions.
• The statistical reliability of virtual ST predictions is currently unclear.

Purpose of the Study:

• To introduce UTOPIA, a model-agnostic framework for multiscale confidence quantification in virtual ST.
• To assign statistically calibrated confidence scores to virtual ST predictions across various spatial resolutions and biological granularities.
• To control false discovery rates for gene and cell type detection in virtual ST.

Main Methods:

• UTOPIA is a model-agnostic framework for confidence quantification.
• It assigns calibrated confidence scores to predictions at different spatial resolutions and biological granularities (genes, metagenes, cell types, cell classes).
• The framework controls false discovery rates while considering local tissue context.

Main Results:

• Prediction confidence in virtual ST is highly dependent on spatial resolution and biological granularity.
• Reliable inference is often achieved at coarser, biologically meaningful scales.
• UTOPIA enhances interpretability and prevents false biological conclusions across multiple ST platforms and settings.

Conclusions:

• UTOPIA provides statistically calibrated confidence scores for virtual ST predictions.
• It enables more trustworthy downstream analyses by ensuring reliable inference.
• The framework is crucial for advancing the application and interpretation of virtual ST methods.