NaS Core is the private research and evidence platform behind the NaS Cortex. It is being built to turn approved biomedical data and literature into reproducible analyses, structured evidence, and traceable research releases.
For the active research flow, its deliverables, and the path from analysis to
patient-relevant decision support, see
CURRENT_PIPELINE.md.
The project has its infrastructure and governance foundations and is now preparing its first governed TCGA-BRCA reproduction study. It does not provide clinical advice and must not be used with protected health information or controlled-access data until the relevant governance controls are approved.
- Python 3.12
- uv
- Docker with Compose (for PostgreSQL and object storage)
cp .env.example .env
uv sync
uv run nas-core storage init
docker compose up -d postgres minio-init
uv run uvicorn nas_core.api.main:app --reloadThe API will be available at http://localhost:8000.
curl http://localhost:8000/health/live
curl http://localhost:8000/health/readymake checkThe authoritative program charter is
workflows/oncology/program_charter.yaml.
It identifies breast oncology as the initial learning domain and explicitly
leaves the first product wedge unselected.
Before collecting articles for a new discovery topic, copy and complete
workflows/templates/research_question_intake.yaml,
score it, obtain review approval, and mark it literature-ready. See
docs/topic-selection.md.
make research-foundation-checkEach approved topic becomes a uniquely identified study workspace with the same
question, literature, protocol, ingestion, analysis, evidence, release, test,
and review stages. Create and validate new workspaces through the CLI; do not
assemble study directories by hand. See workflows/README.md
for the structure, lifecycle gates, and Git-versus-external-storage boundary.
NaS maintains a broad landscape of possible biomedical sources, but every project uses only the minimum sources required by its approved research question. A source appearing below does not mean NaS is authorized to use or store it.
| Source family | Major candidates | What they can support |
|---|---|---|
| Cancer genomics | NCI GDC/TCGA/TARGET, ICGC ARGO, AACR GENIE, cBioPortal, METABRIC | Molecular subtypes, biomarkers, survival, external validation |
| Proteomics and functional biology | NCI PDC/CPTAC, DepMap, CCLE, PRISM, GDSC, LINCS | Protein biomarkers, target discovery, dependencies, drug sensitivity |
| Population oncology | SEER, NCCR, NPCR, NCDB, SEER-linked datasets | Incidence, survival, treatment patterns, disparities |
| Biobanks and longitudinal cohorts | All of Us, UK Biobank, FinnGen, dbGaP, eMERGE, BioVU | Genotype-phenotype, EHR, biomarkers, imaging, longitudinal outcomes |
| Real-world clinical data | Health-system EHRs, registries, claims, labs, pharmacy, pathology, radiology, molecular diagnostics | Treatment response, resistance, toxicity, utilization, patient journeys |
| Imaging and pathology | TCIA, NCI Imaging Data Commons, HTAN, TCGA/CPTAC slides | Radiomics, pathomics, response assessment, multimodal modeling |
| Variants and molecular knowledge | ClinVar, ClinGen, gnomAD, GWAS Catalog, GTEx, COSMIC, CIViC, OncoKB, PharmGKB | Variant annotation, population frequency, gene regulation, clinical evidence |
| Drugs and targets | ChEMBL, PubChem, BindingDB, Open Targets, DrugCentral, DrugBank, Drugs@FDA | Drug-target evidence, bioactivity, pharmacology, approved products |
| Trials and safety | ClinicalTrials.gov, AACT, WHO ICTRP, openFDA, FAERS, FDA/EMA documents | Trial intelligence, eligibility, regulatory evidence, safety signals |
| Literature | PubMed/MEDLINE, PubMed Central, Europe PMC, Crossref, OpenAlex, Cochrane, Embase | Evidence discovery, citation retrieval, review and synthesis |
| Single-cell and spatial | CELLxGENE, Human Cell Atlas, Single Cell Portal, HTAN, HuBMAP | Tumor microenvironment, cell states, spatial biomarkers |
| Proteins and structures | UniProt, RCSB PDB, AlphaFold DB, ESM Atlas, InterPro, STRING, Reactome | Protein function, interactions, structural and mechanistic hypotheses |
The detailed catalog, access tiers, limitations, project-to-source mappings,
and official links are maintained in
docs/data-source-landscape.md.
The distinction is important:
docs/data-source-landscape.mdlists sources NaS may evaluate.data/source-registry.yamlcontains sources actually assessed and approved for defined purposes.- Controlled data and PHI remain prohibited in Cortex v0.
- Open availability never replaces license, attribution, provenance, and project-specific governance review.
Every source follows:
approved research question
-> required modality
-> source assessment
-> access and license review
-> source registration
-> immutable snapshot
-> quality assessment
-> analysis and independent validation
The first pilot plan is
workflows/studies/tcga_brca_stage_survival/protocol/analysis_plan.yaml.
It remains pending scientific review. Validate its structure and data-governance
permissions without downloading study data:
make study-validate
make plan-validatePreview the exact governed request without contacting GDC or writing data:
make gdc-plan-dry-runThe execution path writes content-addressed raw responses and an immutable
dataset manifest to configured object storage. It refuses to run until the
analysis plan has documented founder approval and is marked preregistered. See
src/nas_core/ingestion/README.md for the
operational command and safeguards.
Git contains source code, schemas, workflow definitions, prompts, tests, small synthetic fixtures, and permitted dataset manifests. Raw datasets, credentials, embeddings, generated artifacts, database dumps, PHI, and controlled-access data never belong in Git.
See docs/architecture.md for the initial system shape.
The continuously updated implementation status and next-step queue live in PROJECT_STATUS.md. Read it before beginning material work and update it whenever an implementation is completed.
The complete phased build plan lives in docs/research-engine-plan.md.