Data Sheet 4_SpatialFinder: a human-in-the-loop vision-language framework for prioritizing

Jonathan Xu's SpatialFinder framework combines a biomedical vision-language model with human-in-the-loop optimization to predict gene expression heterogeneity and rank high-value regions of interest (ROIs) across H&E tissue slides. The dataset, last updated in April 2026, contains results from evaluating the framework across four Visium HD tissue types, where it outperformed baseline models for ROI ranking. The framework aims to make spatial transcriptomics more cost-effective and clinically actionable.

Use Cases

• Prioritizing high-value tissue regions for spatial transcriptomics sequencing based on H&E slide analysis.
• Benchmarking vision-language models against human-in-the-loop frameworks for ROI ranking tasks.
• Analyzing gene expression heterogeneity predictions across different tissue types.
• Developing cost-effective workflows for spatial transcriptomics by focusing on smaller, informative regions.

Strengths

• Framework evaluation shows performance gains, with Spearman's ρ up to 0.89 and Overlap@10% up to 78.8%.
• Results are derived from evaluation across four distinct Visium HD tissue types.
• Dataset is shared under a permissive CC-BY-4.0 license.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• The 3.8 MB size suggests the dataset is small and likely contains results or model outputs, not raw sequencing data.

Provenance

Source

figshare, author Jonathan Xu.

Collection Method

Likely contains outputs from the SpatialFinder framework evaluation.

Time Range

Evaluation timeframe is not specified.

Freshness

Last updated 2026-04-15 05:22:58.

Geography

Spatial coverage is not specified.