Spatial Transcriptomics Data for Breast Cancer TLS and HEV Analysis

This dataset contains preprocessed spatial transcriptomics data (AnnData and SpatialData objects), pathology annotations, and cell type mapping results from a study profiling tertiary lymphoid structures (TLSs) and high endothelial venules (HEVs) in breast cancer using three spatial platforms: Visium, Visium HD, and Xenium. Dataset Overview Study Design: 32 patient samples (anonymized as P01-P32) 3 spatial transcriptomics platforms: Visium: 39 samples (32 primary tumors (including 1 technical replicate), 6 matched lymph nodes) Visium HD: 4 samples at multiple resolutions (2µm, 8µm, 16µm, 48µm bins) Xenium: 6 samples at single-cell resolution Naming Convention: Primary tumors: P##_Vis (Visium), P##_HD (Visium HD), P##_Xe5k (Xenium 5K panel) Lymph nodes: P##_LN_Vis Technical replicates: P29_Vis_Rep1, P29_Vis_Rep2 File Structure 1. anndata_objects.zip Preprocessed spatial transcriptomics data stored as AnnData objects (.h5ad format), ready for downstream analysis. These files contain: Gene expression matrices (normalized and log-transformed) Spatial coordinates Low resolution H&E images Contents: visium/ 39 .h5ad files (one per sample) containing Visium spatial data: Spatial resolution: 55µm diameter spots Expression data: ~18,000 genes per sample Includes: Primary tumor samples (P##_Vis) and lymph node samples (P##_LN_Vis) visium_hd/ Visium HD data organized by bin size for 4 samples (P26_HD, P27_HD, P29_HD, P32_HD): square_008um/: 8µm × 8µm square bins square_016um/: 16µm × 16µm square bins square_048um/: 48µm × 48µm square bins (comparable to Visium) Note: The 2µm resolution objects are not included in this record to save space. These objects were not used in the final analysis but can be created from the SpaceRanger outs (see the associated GitHub repo for details). xenium/ 6 .h5ad files (P17_Xe5k, P22_Xe5k, P24_Xe5k, P26_Xe5k, P29_Xe5k, P32_Xe5k): Single-cell resolution spatial data Expression data: 5,100 genes (5K base panel + 100 custom genes) Created from cell-by-gene matrices (not subcellular transcript data) Contains cell centroid coordinates only (segmentation masks not included) Coordinate system matches H&E image 2. pathology_annotations.zip Expert pathologist annotations of tissue compartments at the spot/bin level (CSV format). Each file maps spatial barcodes to tissue annotation labels. Annotation Categories: Epithelial: invasive carcinoma, in situ carcinoma, benign proliferation, normal epithelium Immune: tertiary lymphoid structure (TLS), lymphoid aggregate, lymphocytes present Stroma: desmoplastic stroma, fibrous stroma, adipose tissue Additional features: blood vessel, lymphovascular invasion (LVI), calcification, duct secretion, nerve, necrosis, haemorrhage Contents: visium/ 33 annotation files (one per primary tumor Visium sample): Format: P##_Vis_pathology_annotations.csv Columns include spot barcodes and tissue labels Multi-layer annotations: epithelial, immune, and stromal features annotated separately single_layer column: Collapsed annotation with one label per spot visium_hd/ 4 annotation files for Visium HD samples: Format: P##_HD_pathology_annotations.csv Annotations performed at 8µm bin resolution Single annotation label per bin visium_lymph_nodes/ Annotations for 7 lymph node samples (P12, P25, P26, P27, P28, P29, P32): Two types per sample: P##_LN_Vis_tissue_annotations.csv: Lymphoid tissue area annotations P##_LN_Vis_GC_annotations.csv: Germinal center annotations P##_LN_Vis_GC_annotations.geojson: Germinal center annotations in GeoJSON format (viewable in QuPath) Usage Notes: CSV files can be merged with AnnData objects using barcode/bin IDs as keys GeoJSON files can be visualized in QuPath by opening the corresponding H&E image and importing the annotation file 3. cell_type_mapping.zip Cell type predictions for all spatial samples using single-cell reference-based methods. Results are organized by method and annotation granularity level. Methods: Deconvolution (cell2location): Estimates cell type abundances per spot/bin for Visium and Visium HD Label transfer (Tangram): Assigns cell type labels to individual cells for Xenium Annotation Levels: Each method provides results at three hierarchical levels matching the reference atlas (see original publication for more details: https://doi.org/10.1038/s41588-021-00911-1): major_level/: Broad cell type categories (e.g., T cells) minor_level/: Intermediate cell type subtypes (e.g., CD4+ T cells, CD8+ T cells) subset_level/: Most granular cell type annotations (e.g., CD4+ T-regs, GZMK CD8+ T cells) deconvolution/visium/ Cell type abundance predictions for 33 Visium primary tumor samples: Files per level: P##_Vis_{level}_q05_cell_abundance_w_sf.csv Format: Each row = one spot barcode, columns = cell type abundances Values: Estimated number of cells per cell type per spot (5th percentile of posterior distribution) deconvolution/visium_hd/ Cell type abundance predictions for 4 Visium HD samples: Same structure as Visium deconvolution Performed at two resolutions: 48µm bins (comparable to Visium spots) 16µm bins (higher resolution) Files: P##_HD_{level}_q05_cell_abundance_w_sf.csv label_transfer/xenium/ Cell type predictions for 6 Xenium samples: Files per level: P##_Xe5k_{level}_pred_cell_types.csv Format: Each row = one cell barcode, columns include predicted cell type labels and confidence scores Method: Tangram label transfer maps single-cell reference annotations to Xenium cells based on gene expression similarity Usage Notes: Deconvolution results can be joined to AnnData .obs using spot/bin barcodes Label transfer results can be joined to Xenium AnnData .obs using cell barcodes Choose annotation level based on analysis needs 4. spatialdata_objects.zip SpatialData objects for Xenium samples (.zip format within the main archive). These contain comprehensive spatial data including subcellular transcript positions, cell segmentation boundaries, and metadata - providing access to the full resolution of Xenium data beyond the aggregated cell-level counts in the AnnData objects. Contents: xenium/ 6 SpatialData objects (P17_Xe5k, P22_Xe5k, P24_Xe5k, P26_Xe5k, P29_Xe5k, P32_Xe5k): Transcript-level spatial data with (x, y) coordinates for each detected molecule Cell and nucleus segmentation boundaries (polygon geometries) Cell metadata and expression quantification Coordinates transformed to align with cropped H&E images Enables analysis of subcellular spatial patterns and transcript distributions Note: Morphology images are not included in these objects About SpatialData: SpatialData is a data framework for integrated analysis of multi-modal spatial omics data, designed to handle diverse spatial elements including points (transcripts), shapes (cell boundaries), and images. It provides a unified interface for working with spatial data at multiple scales. Usage Notes: Each sample is stored as a .zip file that must be unzipped before reading. Unpacks into .zarr data. Compatible with Python spatialdata library (use sd.read_zarr() for reading) Complements the cell-level AnnData objects by providing full spatial resolution data