Skip to content

Cancers & treatments

News & evidence

Living with cancer

Learn & contribute

Loading

A public, cited, editable map of oncology: technologies, targets, products, companies, institutions, pathways, trials, pairings, roadmaps, and ideas. One page per object, with a plain-language TL;DR on every page.

Work in progress. Every fact on this site is being built and checked in the open and may be incomplete, out of date, or wrong. You must do your own research and verify anything here at its primary source before relying on it. Nothing on this site is medical advice; decisions belong with you and your clinicians.

OnCo is an information resource, not medical advice. Nothing here replaces a conversation with a qualified clinician. Never delay care because of something you read here; in an emergency, call your local emergency services. Terms of use

Explore
Tags
Search
Path finder
For me
Navigator
Biomarker matrix
Calculators
Interactions
Decision aids
irAE guide
Tumour board
Compare
Landscape grid
Rankings
Timeline
Query
Graph explorer
Saved
Body map

Cancers & treatments

News & evidence

Institutions & people

Institutions
Universities
Trial leadership
People
Heroes and heroines
Companies
Startups
Requests for startups
Investors
Company scorecards
Trial sponsors
Manufacturing map
Countries
India deep dive
China deep dive
Cases by country
Funding flows

Living with cancer

For me
The first 60 days
Appointment prep pack
Side effects
Symptom to test
Report reader
Survivorship planner
Second opinion
Financial help
Free in oncology
Paying for care in the US
NHS coverage
Paying for care in India
Coverage rankings
Cutting cancer care costs
Diet, exercise and lifestyle
Supportive care
Complementary and supportive approaches
Hair loss and regrowth
Heroes and heroines

Learn & contribute

Reading paths
About & methodology
Roadmap
Gaps to fill
Completeness
Review queue
Reviewer roster
Contributors
Idea votes
Teaching packs
Weekly issue
Suggest an edit
Open evaluation
Data dictionary
Build on OnCo
Open API
Open data
Open tools
Open source in oncology
GitHub

© 2026 OnCo. Code MIT. The data are free for individual and educational use with the attribution “Data from OnCo (onco.cc)”, under CC BY-NC 4.0. Commercial use must contact us to pay for the data. Commercial licences. Made with heart and soul by Jude Gomila .

About and methodology Corrections Open API Tags Terms of use Privacy policy GitHub

Loading

Home
Glossary
Cancer AI vocabulary (CanSim terms map)

Cancer AI vocabulary (CanSim terms map)

aka cancer AI vocabulary, CanSim terms map, CanSim terms, methods and models glossary

A hub for the vocabulary of cancer AI: the assays and cohorts models train on, the machine-learning and statistics terms in their papers, the standards their data must follow and the licences that govern reuse.

Overview Notes Connected150

Overview

OnCo added this vocabulary on 24 September 2026 from the CanSim terms map, an open list of the terms an open, public-data-first cancer foundation-model programme found itself explaining. The oncology half (assays, data modalities, standards, access rules, clinical concepts) sits in the canonical glossary categories; the methods half sits under Methods and models. Every term paraphrases the page it links, states why a reader meets it in cancer papers, and carries the attribution below. Terms that were only CanSim result names with no general meaning were left out and are listed in docs/CANCERSIM-TERMS-GAP.md. Attribution: CanSim terms map 1.0.0 (docs/onco/terms.json, generated 2026-09-24), CC BY 4.0, attribution: CanSim project, an open, public-data-first cancer foundation-model programme.

Stained protein (HER2, PD-L1)
Glass slide
Whole-slide scan → model

Schematic · not to scale

Stained protein (HER2, PD-L1) · Glass slide · Whole-slide scan → model

Showing the technology this term belongs to: Pathology & radiology foundation models.

Category

Methods and models

Notes

Research vocabulary, not clinical advice: models and metrics described here are research tools unless a page says a product is cleared or approved.

Connected

technologies

5

AI compute and model platforms for oncology Digital pathology & AI Pathology & radiology foundation models RNA sequencing & expression profiling Single-cell & spatial profiling

targets

11

DCLK1 DNAJC12 E2F7 GRB7 MIEN1 PGAP3 PHLDA1 PHLPP2 PNMT SPRY4 STARD3

terms

128

Ablation study and multi-task heads Accuracy, macro-F1 and confusion matrices Actionable genomic biomarkers AJCC stage (TNM staging manual)Analytical databases as a data catalogue (DuckDB, PostgreSQL)Analytical versus clinical validation AnnData and h5ad files Array and table formats: HDF5, Zarr, OME-Zarr, Parquet Attention-based multiple-instance learning (ABMIL, CLAM)Autoregressive (next-token) modelling Batch effects and harmonisation Benchmarks, leaderboards and contamination Bootstrap resampling Bulk RNA sequencing (RNA-seq) and the full transcriptome C-index (concordance index), Harrell's and Uno's Calibration: reliability diagrams and the Brier score Cell composition confound (tumour versus stroma and immune cells)Cell lines as a proxy for patients Censoring and events in survival data CITATION.cff (Citation File Format)Clinical covariates (age, stage, nodes, treatment flags)Clinical text: EHR notes and pathology reports Co-amplification and the 17q12 HER2 amplicon Conformal prediction Contrastive learning (InfoNCE)Controlled-access genomic data (dbGaP, EGA)Copy number alteration (CNA)Cross-validation and stratified k-fold Data leakage in model evaluation Data use agreements and consent for research use Deep mass-spectrometry proteome (CPTAC)Digital pathology and whole-slide images (WSI)DNA methylation arrays and beta values (450k, EPIC)Domain shift and domain adaptation (cell line to patient)Drug response and drug sensitivity (IC50, AUC)Drug-response baselines and frameworks: mean-drug floor, LightGBM, DrEval, IMPROVE, DeepTTA Drug-response data splits: leave-cell-line-out, leave-drug-out, leave-tissue-out Embedding (learned representation)Endocrine therapy and endocrine resistance Ensembl gene ID External validation Fine-tuning versus frozen features, and LoRA Foundation model Gene co-expression structure Gene set enrichment analysis (GSEA and ssGSEA)Genome builds: GRCh38 versus hg19 (GRCh37)Genomic and protein language models: Evo 2, Enformer, ESM GISTIC (copy number driver detection)and 80 more →

ideas

2

An open foundation model of the cancer cell trained on perturbation data Patient-level multimodal foundation models for treatment selection

collections

4

CPTAC (Clinical Proteomic Tumor Analysis Consortium)CZ CELLxGENE / Human Cell Atlas DepMap (Cancer Dependency Map)TCGA / NCI Genomic Data Commons

Not yet reviewedNeeds patient advocate review. Review this page or see the review queue and roster.

Model panel commentary is coming to this page: machine commentary, not clinical review.

cansim-terms hub

Data

JSON · ·

Follow this page

Wrong or missing?

Propose a change with a source. Organisations can update their own records. Every suggestion is reviewed, safety-checked and validated before it goes live; nothing is edited directly.

Suggest an editDiscuss start a thread

Out of date?
How review works

Technologies

Pathology & radiology foundation models
AI compute and model platforms for oncology
RNA sequencing & expression profiling
Single-cell & spatial profiling
Digital pathology & AI

Terms

Tumour board (multidisciplinary team meeting)
Grade versus stage
Tumour purity
Cell composition confound (tumour versus stroma and immune cells)
Intra-tumour heterogeneity
Tumour evolution (somatic evolution)
Censoring and events in survival data
Risk score and risk percentile
Clinical covariates (age, stage, nodes, treatment flags)
Endocrine therapy and endocrine resistance
Immunotherapy response and its prediction
Tertiary lymphoid structures (TLS)
Pharmacogenomics
Drug response and drug sensitivity (IC50, AUC)
Cell lines as a proxy for patients
Co-amplification and the 17q12 HER2 amplicon
Copy number alteration (CNA)
Actionable genomic biomarkers
mRNA to protein concordance
Post-transcriptional and post-translational regulation
Pathway activation state (phosphosignalling)
Gene co-expression structure
Tissue-of-origin signal in tumour data
Spatial autocorrelation (Moran's I)
Variant effect prediction
Bulk RNA sequencing (RNA-seq) and the full transcriptome
TPM, FPKM and raw counts (expression units)
STAR and Salmon (RNA-seq alignment and quantification)
Variant calling
Somatic mutations from exome and genome sequencing (WXS, WGS)
Targeted panel sequencing
GISTIC (copy number driver detection)
MutSig (significantly mutated gene detection)
Batch effects and harmonisation
Gene set enrichment analysis (GSEA and ssGSEA)
Single-cell RNA sequencing (scRNA-seq, 10x Chromium)
Spatial transcriptomics platforms (Visium HD, Xenium, MERFISH, CosMx, CODEX)
H&E staining (haematoxylin and eosin)
Digital pathology and whole-slide images (WSI)
Tile and patch encoding of slides
Magnification (20x, 40x) and microns per pixel
DNA methylation arrays and beta values (450k, EPIC)
Microarray expression data
Reverse-phase protein array (RPPA)
Deep mass-spectrometry proteome (CPTAC)
Clinical text: EHR notes and pathology reports
Radiology imaging as a data modality (CT, MRI, TCIA)
Survival outcomes as time plus event (OS, PFS)
Controlled-access genomic data (dbGaP, EGA)
TCGA open versus controlled data tiers
Data use agreements and consent for research use
Software as a medical device (SaMD)
Research use only (RUO)
Analytical versus clinical validation
HGNC gene symbol
Ensembl gene ID
Genome builds: GRCh38 versus hg19 (GRCh37)
HGVS variant nomenclature
ICD-O-3 (International Classification of Diseases for Oncology)
OncoTree cancer classification
NCI Thesaurus (NCIt)
Mondo disease ontology
Human Phenotype Ontology (HPO)
Uberon anatomy ontology and the Cell Ontology
Units of measurement ontology (UO)
RxNorm drug nomenclature
AJCC stage (TNM staging manual)
TCGA barcode
Provenance fields for research data
Stroma-rich and desmoplastic tumours in molecular data
Foundation model
Self-supervised pretraining (SSL)
Masked autoencoders and masked gene modelling
Contrastive learning (InfoNCE)
Transformer and attention
Tokenisation (genes, tiles and sequence as tokens)
Embedding (learned representation)
Fine-tuning versus frozen features, and LoRA
Linear probe
Transfer learning and the low-label regime
Zero-shot prediction
Autoregressive (next-token) modelling
Multimodal fusion (early, late, modality dropout)
Attention-based multiple-instance learning (ABMIL, CLAM)
Pathology foundation models: UNI, UNI2, Virchow2, CTransPath, CONCH, TITAN
Single-cell and transcriptome foundation models: UCE, GeneCompass, BulkFormer, BulkRNABert
Genomic and protein language models: Evo 2, Enformer, ESM
Spatially aware clustering (SpaGCN, KNN smoothing)
Virtual cell models and in-silico perturbation screens
Drug-response data splits: leave-cell-line-out, leave-drug-out, leave-tissue-out
Drug-response baselines and frameworks: mean-drug floor, LightGBM, DrEval, IMPROVE, DeepTTA
Domain shift and domain adaptation (cell line to patient)
Mechanism-of-action recovery and known-biology probes
Cross-validation and stratified k-fold
Train, validation and test split discipline
Data leakage in model evaluation
External validation
C-index (concordance index), Harrell's and Uno's
Calibration: reliability diagrams and the Brier score
ROC-AUC, PR-AUC and time-dependent AUC
Accuracy, macro-F1 and confusion matrices
Univariate Cox scores and time-dependent metrics
Bootstrap resampling
Permutation test and the Mann-Whitney U test
Spearman rank correlation
Logistic regression and nearest-centroid classifiers
Ridge regression and the multilayer perceptron
Principal component analysis (PCA) as a feature compressor
Quantile normalisation, rank transforms and z-scores
Loss functions: cross-entropy and mean squared error
Ablation study and multi-task heads
Out-of-distribution detection (Mahalanobis guard)
Uncertainty quantification and confidence gates
Conformal prediction
Model cards and datasheets for datasets
Benchmarks, leaderboards and contamination
Reproducibility and negative results
Pre-registered experiment
Open weights, open code and gated models
Open licences: Apache-2.0, MIT and CC BY 4.0
Hugging Face Hub
Zenodo DOIs for data and code
CITATION.cff (Citation File Format)
AnnData and h5ad files
Array and table formats: HDF5, Zarr, OME-Zarr, Parquet
Analytical databases as a data catalogue (DuckDB, PostgreSQL)
GPU training and mixed precision
Local-only language models over patient data (privacy by architecture)

Related

Patient-level multimodal foundation models for treatment selection
An open foundation model of the cancer cell trained on perturbation data
TCGA / NCI Genomic Data Commons
CPTAC (Clinical Proteomic Tumor Analysis Consortium)
DepMap (Cancer Dependency Map)
CZ CELLxGENE / Human Cell Atlas

Machine-readable versions of Cancer AI vocabulary (CanSim terms map)

JSON record with neighbours
Markdown context for language models
RDF Turtle for this record (its triples, relations and owl:sameAs; the IRI of this record is https://onco.cc/terms/cancer-ai-vocabulary/)
All terms as JSON
RDF N-Triples for the whole corpus
Search index: id, kind, name, TL;DR and route for every record
OpenAPI 3.1 description
API root and meta.json
llms.txt

MCP server: npx -y onco-mcp, tools search, get_entity, list_kind, ask, context, compare. This record is get_entity or context with id cancer-ai-vocabulary.

WebMCP tools registered on this page: onco_search, onco_get_entity.