This video demonstrates the core functionality of the SpotWhisperer web application. It showcases how users can load a histopathology sample, perform natural language queries to identify cellular regions like "T cells," and explore the underlying inferred gene expression data that drives these predictions. The demonstration highlights the intuitive, interactive, and molecularly-grounded approach of our method.

Spot level H&E annotation benchmark

This benchmark dataset consists of hematoxylin and eosin (H&E) images from five lung cancer samples (Dawo et al., 2025), annotated with cell types and pathologically relevant labels at spot-level resolution. For methodology, refer to the accompanying publication (Schaefer et al., 2025)

Dataset Files

The dataset contains the following compressed h5ad files:

• LC1.h5ad.gz - Lung cancer sample 1
• LC2.h5ad.gz - Lung cancer sample 2
• LC3.h5ad.gz - Lung cancer sample 3
• LC4.h5ad.gz - Lung cancer sample 4
• LC5.h5ad.gz - Lung cancer sample 5

and can also be downloaded as a complete (compressed) tar archive.

---

Usage Instructions

Loading and Basic Analysis

import scanpy as sc

# Read the curated file
file_path = 'LC1.h5ad.gz'
adata = sc.read_h5ad(file_path)

# Display basic information
print(adata)

# Information about spots
print(adata.obs)

# Information about genes
print(adata.var)

# High-resolution image
print (adata.uns["20x_slide"])

# Sample metadata
print (adata.uns["meta"])

# Spatial coordinates
print (adata.obsm["X_spatial"])

---

Key Data Components

Below is a summary of the essential data components for the curated data:

Gene Expression Data

• adata.X contains raw gene expression counts (SPOTS × GENES)

Gene Information

• adata.var contains human-readable gene names, gene IDs, and feature_types

Spatial Information

• Spot coordinates:
  • adata.obs['x_array'], adata.obs['y_array']: Original spot coordinates on the slide space
  • adata.obs['x_pixel'], adata.obs['y_pixel']: Spot coordinates in the image space
  • adata.obsm['X_spatial']: 2D spatial coordinates for visualization

High-resolution image

Image

• adata.uns["20x_slide"]: High-resolution image
• adata.uns["meta"]["magnification"]: Magnification of the image

Annotations and Metadata

• Tissue coverage:
  • adata.obs['in_tissue']: Boolean indicating spots within tissue boundaries as annotated by histopathologists
• Expert annotations:
  • adata.obs['region_type_expert_annotation']: Manual tissue region annotations
  • Categories: UNASSIGNED, NOR (Normal), TUM (Tumor), TLS (Tertiary Lymphoid Structures), INFL (Inflammatory)
• Cell type annotations:
  • adata.obs['cell_type_annotations']: Automated cell type predictions from reference atlas
  • Provides broad cell type classifications (e.g., Epithelial, Immune, Stromal)
• Sample metadata:
  • adata.obs['sample_ID']: Sample identifiers
  • adata.obs['barcode']: Unique spot barcodes
  • adata.uns['meta']: Sample metadata
  • adata.uns['meta']['spot_diameter_fullres']: Spot diameter
  • adata.uns['meta']['dot_size']: Number of subspots within a spot (DeepSpot inference parameter)