SciGroveBeta
Genetics

COAST: Context-Aware Differential Learning for Gene Expression Prediction in Spatial Transcriptomics

Keunho Byeon, Sunhong Park, Jeewoo Lim, Jin Tae Kwak

Featured July 14, 2026

Original Paper

This analysis was generated by SciGrove. Upload your own PDFs or enter a DOI — and get the same AI breakdown on any paper.

Get started

AI-generated analysis — This is SciGrove's AI interpretation of the paper, not peer-reviewed content. Always refer to the original paper.

Simply

A new method predicts gene activity in tissue by looking at both the specific spot and its surroundings, learning not just how much gene activity there is, but also how it changes relative to nearby and distant areas.

In depth
The paper introduces COAST, a framework that predicts spatial gene expression from H&E images by explicitly leveraging spatial context. It achieves this by adaptively modulating local and global context features based on their type and aggregating them with target spot features using a Transformer. A key innovation is a joint objective function that supervises both absolute gene expression and the *signed differential expression* between a target spot and its context, making predictions more robust to slide-level variability.

Key Takeaways

  • 1
    The framework integrates local and global spatial context from H&E images to predict gene expression, moving beyond isolated patch-level analysis.
  • 2
    A novel differential learning objective explicitly supervises relative gene expression changes between spots, enhancing robustness and capturing meaningful spatial gradients.
  • 3
    Context-specific feature modulation and Transformer-based aggregation allow the model to adaptively process heterogeneous spatial information.

Conceptual Flow

HIGH LEVEL
1
Methodology: Learning from Context and Differences

The method takes a tissue image, finds features for a spot and its neighbors, then combines these features to predict gene activity, learning both the exact amount and how it differs from nearby spots.

Tissue Image
Extract Features
Spot Features
Neighbor Features
2
Results: Improved Prediction and Clinical Relevance

The new method predicts gene activity more accurately than previous ways and helps identify important patterns in patient survival.

Old Prediction
New Prediction
Compare Accuracy
New Method Better