Downloads
Every table on this site as a tab-separated file, ready for R, Python or a spreadsheet. If you use them, please cite LB2 and the underlying datasets.
Candidate tables
Every candidate with both comparisons, confidence intervals, q-values, immune-cell expression and plasma evidence.
All cancer types in one table
48,527 rows from 29 cancer types, built in your browser (about 9 MB).
| Code | Cancer type | Candidates | File |
|---|---|---|---|
| ACC | Adrenocortical carcinoma | 548 | |
| BLCA | Bladder urothelial carcinoma | 680 | |
| LGG | Brain lower grade glioma | 3,084 | |
| BRCA | Breast invasive carcinoma | 590 | |
| CESC | Cervical squamous cell carcinoma and endocervical adenocarcinoma | 2,574 | |
| CHOL | Cholangiocarcinoma | 4,915 | |
| COAD | Colon adenocarcinoma | 1,356 | |
| ESCA | Esophageal carcinoma | 2,246 | |
| GBM | Glioblastoma multiforme | 1,816 | |
| HNSC | Head and neck squamous cell carcinoma | 1,463 | |
| KICH | Kidney chromophobe | 343 | |
| KIRC | Kidney renal clear cell carcinoma | 1,267 | |
| KIRP | Kidney renal papillary cell carcinoma | 553 | |
| LIHC | Liver hepatocellular carcinoma | 1,626 | |
| LUAD | Lung adenocarcinoma | 1,364 | |
| LUSC | Lung squamous cell carcinoma | 2,156 | |
| MESO | Mesothelioma | 1,378 | |
| OV | Ovarian serous cystadenocarcinoma | 3,084 | |
| PAAD | Pancreatic adenocarcinoma | 1,136 | |
| PCPG | Pheochromocytoma and paraganglioma | 3,553 | |
| PRAD | Prostate adenocarcinoma | 649 | |
| READ | Rectum adenocarcinoma | 1,355 | |
| SARC | Sarcoma | 1,981 | |
| SKCM | Skin cutaneous melanoma | 1,363 | |
| STAD | Stomach adenocarcinoma | 2,584 | |
| TGCT | Testicular germ cell tumors | 1,165 | |
| THCA | Thyroid carcinoma | 685 | |
| UCS | Uterine carcinosarcoma | 2,183 | |
| UCEC | Uterine corpus endometrial carcinoma | 830 |
Columns in the candidate tables
- tcga_code, cancer_type
- TCGA project code and study name
- rank
- position by priority score within the cancer type
- symbol, ensembl_gene_id
- HGNC symbol and versioned Ensembl ID (GENCODE v23)
- gene_type, gene_class
- GENCODE v23 biotype and its coarse class (protein-coding, lncRNA, pseudogene, small RNA, other)
- plasma_status
- evidence, not_detected (protein-coding, in no source) or not_eligible (non-coding)
- auc_vs_normal, …_ci_low, …_ci_high
- AUC against normal tissue with its DeLong 95% confidence interval
- log2fc_vs_normal, fdr_vs_normal
- difference in mean log2(TPM + 0.001) and Benjamini–Hochberg q-value
- auc_vs_blood, … fdr_vs_blood
- the same measures against GTEx whole blood
- immune_max_ntpm
- highest nTPM across 18 sorted immune cell types; empty if the gene is not in the HPA reference
- plasma_n_sources, plasma_sources
- number and names of plasma sources that support the gene
- in_tissue_panel, panel_coef
- whether the gene is in the tissue panel, and its coefficient
- priority_score
- mean of the two AUCs plus 0.05 per plasma source (at most 0.15)
Summary tables
- Cancer type summaryTSV, 5 KB
One row per cancer type: sample sizes by source, genes remaining after each selection step, panel size and held-out AUC, survival endpoint and events.
lb2_cancer_summary.tsv
- Tissue classifier panelsTSV, 150 KB
Every gene selected by the elastic-net panels, with its standardized coefficient, the panel's held-out AUC, and whether the expression was study-corrected first.
lb2_tissue_panels.tsv
- Survival modelsTSV, 454 KB
All Cox models, including non-significant ones: hazard ratio per SD with 95% CI, p-value and FDR for the 150 top-ranked candidates of each cancer type.
lb2_survival_cox.tsv
- Shared NSCLC candidatesTSV, 255 KB
Genes that are candidates in both lung adenocarcinoma and lung squamous cell carcinoma, with each project's statistics side by side.
lb2_shared_nsclc.tsv
- Shared colorectal candidatesTSV, 280 KB
Genes that are candidates in both colon and rectal adenocarcinoma (primary tumors), with each project's statistics side by side.
lb2_shared_crc.tsv
Programmatic access
The site reads static JSON you can fetch directly: /data/atlas.json for the overview and /data/cancer/<slug>.json for each cancer type, with candidates stored column by column.
For custom queries, run the pipeline and its FastAPI service from the repository.
Terms
LB2’s code is MIT-licensed. The derived tables build on TCGA, GTEx, the Human Protein Atlas, exoRBase and GEO, and inherit their terms; check them before redistributing. Results are for research only.