OpenTME: Open-Access Spatial Tumor Microenvironment Profiles from TCGA
OpenTME, released by Aignostics, is a gated, open‑access dataset that delivers pre‑computed spatial readouts for whole‑slide H&E images from The Cancer Genome Atlas (TCGA). The data are derived from the Atlas H&E‑TME foundation‑model pipeline, which performs tissue quality control, segmentation, cell detection/classification, and neighborhood analysis. For each slide, CSV files provide quantitative metrics on image quality, tissue composition, cell type counts and densities, and spatial co‑occurrence within 20 µm and 40 µm neighborhoods.
The current release (v1.1, June 2026) covers eight primary cancer sites—breast, bladder, colorectal, liver, lung, pancreas, prostate, and stomach—totaling over 5,000 WSIs. The dataset falls in the 10K < n < 100K size category and is tagged for image‑classification, image‑segmentation, feature‑extraction, and object‑detection tasks, reflecting the underlying computational pathology workflow that generated the metrics. Researchers can access the data after completing a short gated form that requires an academic or non‑profit email address.
OpenTME is intended strictly for non‑commercial academic research. Its license forbids commercial use, redistribution, and any AI/ML training that attempts to replicate the Atlas H&E‑TME analysis. Users are encouraged to cite the associated arXiv preprint (arXiv:2604.12075) and to respect TCGA data‑use policies. Aignostics also provides the TME Studio Marimo notebooks on GitHub to help users load, filter, and explore the CSV readouts.
Project Ideas
- Perform comparative analysis of cell type composition across the eight cancer types to identify tissue‑specific immune signatures.
- Correlate tissue‑level metrics (e.g., stroma area, necrosis fraction) with publicly available TCGA clinical outcomes for survival or treatment response studies.
- Build an interactive dashboard that visualizes QC, tissue, and cell metrics per slide, enabling researchers to filter by cancer type and metric thresholds.
- Apply clustering algorithms to the neighborhood metrics to discover distinct tumor microenvironment phenotypes within and across cancer sites.
- Integrate OpenTME CSV data with external genomic datasets (e.g., mutation burden) to explore associations between spatial microenvironment features and molecular alterations.