Resource guide
Guide
Learn how to browse the atlas, interpret a virtual spatial map, run prediction on your own H&E image, and report the outputs responsibly.
Browse the atlas
The atlas contains 6,021 indexed TCGA whole-slide predictions across BRCA, SKCM, and COAD/READ.
- Search or filter
Use a cohort name, case ID, slide ID, or gene keyword. Filters can restrict the table by cohort, spatial-map availability, H&E availability, and tile count.
- Open one slide
Select a slide ID in the table to enter its interactive record. The default order groups BRCA, SKCM, and COAD/READ and ranks slides by tissue-tile count within each cohort.
- Export the current view
Use Export Current View to save the current filtered slide manifest for reproducible downstream selection.
Read a virtual spatial map
Slide Detail combines the H&E thumbnail with tile-level HistoOmniST predictions.
- Choose a feature
Select a gene or spatial program from the control bar. The map redraws the selected virtual signal without changing the underlying H&E image.
- Adjust point size
Point size changes only the display. Hover over a point to read its tile coordinates, predicted value, and tissue fraction.
- Read the color scale
Colors are scaled within each slide, usually after percentile clipping. Use them to inspect within-slide spatial patterns, not to compare absolute intensity between slides.
Run prediction on your image
The upload workflow accepts SVS, TIFF, PNG, and JPG images and stores each submission in a private job workspace.
- Select an image
Upload one or more H&E images. The default dense mode is intended for formal spatial-map inspection.
- Review advanced settings only when needed
Target spots, spot limit, and minimum stride control spatial sampling density and runtime. They do not change the trained model parameters.
- Open the completed result
The result page provides an H&E thumbnail, interactive gene and program maps, state maps, prediction tables, an integrated report, and a download package.
Understand the outputs
The website exposes model-derived virtual signals at tile or spot resolution.
| Output | Meaning | Recommended use |
|---|---|---|
gene_* | Count-log1p virtual expression reconstructed from rate and mean-one predicted SF | Within-slide gene-pattern inspection |
count_* | Count-scale virtual signal reconstructed with predicted SF | Count-scale analyses with the stated model version |
pred_sf | Predicted slide-normalized size factor | Rate-to-count reconstruction and QC |
program_* | Predicted spatial-program score | Inspect epithelial, stromal, immune, and proliferation patterns |
state map | Tile state assigned from predicted program profiles | Summarize recurrent tissue states |
Interpret results responsibly
Virtual spatial transcriptomes are computational predictions and must be reported with their validation scope.
TCGA maps are model-derived signals, not measured spot-level spatial transcriptomes.
HEST validation shows stronger and weaker tissues. Do not generalize a strong organ result to every tissue.
Survival and molecular associations support biological relevance but are not deployed diagnostic or treatment-decision systems.
Privacy, citation, and reuse
Uploaded jobs and downloaded predictions require clear provenance and responsible handling.
For security, abuse prevention, and service reliability, the server records source IP, request time, path without query parameters, status code, request duration, and browser identifier. Raw logs are retained for 10 days and are not published as atlas data.
Uploaded jobs are accessed through a private URL or browser-stored token. Public pages do not list other users' submissions.
Cite HistoOmniST and the original cohort or validation dataset. Describe downloaded values as model-derived virtual spatial transcriptome signals.
Report the model version, cohort, slide ID, selected gene or program, spatial sampling settings, and color-scale convention.