DNA / WGS
Whole genome
Broadest view of inherited and tumor-acquired change.
- SNVs and indels
- Copy number and structural variation
- Non-coding and breakpoint context
- Best paired with matched normal
The data foundation
Collect once. Clean once. Analyze many times.
Every contribution should enter the Foundation's data as a complete, longitudinal, analysis-ready patient record, not as another folder of disconnected files.
This page is a design. It describes how the Foundation intends to collect, clean, govern and share the data it will hold. None of it is running today, and the legal statements in it are drafts for counsel and IRB review.
- All modalities connected
- Every timepoint preserved
- Source data never overwritten
- Harmonized at contribution
Arrives as
FASTQ / BAM / VCFDNA, RNA, long-read
Reports / spreadsheetsLab calls and interpretation
Clinical / imagingTreatment, toxicity, outcome
Consent + use rightsFHIR Consent · DUO terms · tier · version
Harmonized
at
contribution
at
contribution
Leaves as
Longitudinal patient graphEvery sample and exposure linked
Analysis-ready matricesConsistent coordinates and vocabularies
Queryable, with the consent attachedAsk the cohort question first
No file hunting. No manual sample matching. No coordinate liftover. No bespoke ETL.
~250files in one illustrative longitudinal patient catalog
~559 GBraw and lab-processed molecular data
5source folders with different lab conventions
5+serial tissue timepoints across treatment states
7+molecular and functional modalities represented or planned
Private patient identifiers and exact dates are omitted from this public page.
01 / Complete patient record
What data is critical
The biological data only become useful when they remain attached to the patient journey.
The unit of analysis is not a sequence file. It is a patient, a disease state, a lesion, a specimen, a treatment exposure, and an outcome, linked over time.
Clinical timeline + treatment exposure
Imaging, pathology + lesion response
ctDNA, labs, symptoms + outcomes
DNA: WGS, WES, targeted panels
The patient
one continuous record
RNA: short-read + long-read
Proteome + phosphoproteome
Single-cell + spatial biology
Immunopeptidome + immune state
Clinical context
Diagnosis, disease burden, performance, comorbidities, every treatment, dose, hold, surgery, radiation, response and toxicity.
Biospecimen context
Lesion identity, anatomic site, collection time, preservation, tumor content, necrosis, cold ischemia and pathology review.
Raw molecular data
Instrument-level files whenever available, retained immutably so future methods can reprocess the original evidence.
Lab-derived results
Original reports, spreadsheets, calls and interpretations are preserved beside the harmonized analysis, not substituted for it.
Longitudinal outcomes
Early molecular response, imaging, symptoms, quality of life, progression, resistance and subsequent interventions.
Complete provenance
Who generated each file, with which instrument, reference, software, parameters, annotation set and quality status.
02 / Complementary sequencing
Different tests answer different questions
WGS, WES, RNA, long-read and single-cell data are not interchangeable.
The record should preserve the strengths and limitations of each assay, then connect them into a coherent view rather than flattening them into one vendor report.
DNA / WES
Whole exome
Efficient depth across protein-coding genes and a strong cross-validation layer.
- Actionable coding variants
- Tumor-normal comparison
- Serial clonal change
- Capture-kit effects retained
DNA / Targeted
Clinical panels
Useful clinical snapshots with fast interpretation, but limited genomic breadth.
- Hotspots and selected genes
- Clinical lab assertions
- Panel-covered negatives only
- Never treated as equivalent to WGS
RNA / Short-read
Transcriptome
Shows which genes and pathways are active at a particular disease state.
- Expression counts and TPM
- Fusion support
- Immune and pathway signals
- Allele-specific expression
RNA / Long-read
Full-length isoforms
Resolves transcript structure that short reads can fragment or misassign.
- Isoforms and splice junctions
- Fusion transcript architecture
- Novel transcript discovery
- Cross-check against short-read RNA
Cells / Spatial
Heterogeneity
Separates tumor, immune and stromal states that bulk measurements average together.
- Cell populations and states
- Rare resistant subclones
- Cell-cell interactions
- Spatial organization when available
One timepoint is a snapshot. Repeated sequencing turns the disease into a movie.
03 / Longitudinal by design
Every specimen is anchored to treatment
Track what changed, when it changed, and what pressure selected it.
Serial tissue and blood allow within-patient comparisons across diagnosis, treatment, recurrence and metastasis. Multiple lesions collected at the same procedure reveal spatial heterogeneity as well as temporal evolution. The timepoints below are from the illustrative catalog.
- 01T0
Pre-treatment biopsy
Baseline biology before systemic therapy.
WESWTSpanel - 02T1
Resection after initial therapy
Residual disease after first treatment pressure.
WESWTSpathology - 03T2
Recurrence biopsy
Biology after completion of standard therapy.
WESWTSreport - 04T3
Resection after salvage exposure
Fresh-frozen and FFPE material for deeper analysis.
WGS*WESWTSlong RNAprotein - A+BT4
Two lung metastases
Independent nodules with shared matched normal.
WESWTSproteinimmune - nextT5
The next resection
Repeat the core assays and add high-value modalities where tissue permits.
WGSWTSlong RNAsingle-cellfunctional
*An expected WGS component in the illustrative catalog was not present in the delivered folder. The record would mark it expected-but-missing, not as a negative result.
04 / One sample lineage
A graph, not a file cabinet
Every result must be traceable back to the exact patient, lesion, specimen and assay.
Stable identifiers and validated relationships prevent the most damaging class of errors: mispaired normals, merged lesions, ambiguous timepoints, lost aliquot history and results detached from treatment context.
Patient
Disease episode
Procedure
Lesion
Specimen
Exact treatment state and time since exposure, linked to specimen collection
Aliquot / analyte
Assay run
Raw file
Harmonized file
Derived feature
Every node has a durable ID.
The graph keeps biological replicates, technical replicates, multiple nodules, shared normals and repeated assays distinct while making their relationships directly queryable.
A file without lineage is stored, but it is not labeled analysis-ready.
05 / Clean at contribution
The work researchers should not have to repeat
Ingest, validate, align, normalize and document the data once.
A contribution would count as complete only after it passes a data contract and produces versioned, quality-scored, analysis-ready outputs.
- 1
Manifest + integrity
Inventory every expected and received file; verify checksums, byte size, compression and readability.
- 2
Identity + pairing
Fingerprint samples, confirm patient identity, tumor-normal pairing, lesion IDs, sex and contamination.
- 3
Metadata contract
Map site, timepoint, preservation, assay, platform, panel, read group and treatment context to controlled terms.
- 4
Technical QC
Coverage, duplication, insert size, mapping, RNA quality, tumor purity, read structure and modality-specific checks.
- 5
Canonical alignment
Reprocess DNA and RNA against a versioned common reference and annotation bundle while retaining source alignments.
- 6
Standard calling
Generate SNV/indel, CNV, SV, fusion, expression, isoform and other derived outputs with reproducible workflows.
- 7
Normalize + annotate
Normalize coordinates and identifiers; add transcript, gene, protein, pathway and clinical annotations with versions.
- 8
Reconcile platforms
Measure concordance, retain disagreement, identify report-only evidence and never convert missing into negative.
- 9
Privacy + use rights
Apply de-identification, access tier, consent and data-use tags. Preserve intervals in longitudinal research views. Section 06 describes the machine-readable stewardship layer.
- 10
Publish + refresh
Release immutable raw, harmonized and analysis-ready layers through versioned snapshots.
| Data type | Work completed at contribution | Research-ready result |
|---|---|---|
| DNA | Common reference; read-group validation; tumor-normal identity; SNV/indel, copy number and structural-variant processing; variant normalization. | Comparable variant and copy-number tables across laboratories, panels and timepoints, plus source-native calls. |
| Short-read RNA | Common genome and transcript annotation; gene-level counts; expression normalization; fusion analysis; quality and batch covariates. | Counts, TPM, fusions, pathway-ready matrices and explicit assay/batch metadata. |
| Long-read RNA | HiFi QC; transcript alignment; isoform collapse and classification; junction and fusion reconciliation with short reads. | Full-length isoform and transcript-event tables linked to expression and genomic breakpoints. |
| Single-cell / spatial | Barcode and UMI QC; doublet and ambient-RNA assessment; raw-count preservation; cell and region annotations; batch flags. | Portable matrices with curated cell labels, embeddings, QC fields and spatial coordinates where available. |
| Protein / peptide | Standard protein and phosphosite IDs; FDR and intensity normalization; peptide-to-protein and variant-peptide mapping; raw-spectra status. | Comparable protein, phosphosite and immunopeptide features with confidence and provenance. |
| Clinical / imaging | Medication, lab, disease, toxicity and response vocabularies; lesion IDs; DICOM de-identification; treatment intervals and outcomes. | Queryable treatment-exposure, lesion-response, toxicity and outcome tables synchronized to molecular timepoints. |
06 / Consent once
Consent once. Contribute everywhere.
One patient choice should travel with every specimen, file and derived result.
The design is one portable consent and stewardship layer for trials, institutions and patient-mediated contribution. Every future trial would contribute through that same layer. Each patient would choose how their data may be used, and every permission would become a machine-readable rule that follows the longitudinal patient graph.
Concept draft. The legal and regulatory statements in this section are for counsel and IRB review. Portability depends on IRB approval, reliance or contribution agreements, and the privacy law of each jurisdiction. The full legal basis table is in the working specification, which is not published here.
Patient
Chooses scopes and separate permissions
Portable source of truth
Patient consent
Stored as an HL7 FHIR Consent resource
Versioned · effective dates · withdrawal state
Every graph node
Data-use tags
DUO terms plus consent-profile fields
PurposeTierEmbargoJurisdiction
Open aggregateCohort summaries. No individual records. Always available.
Treating teamCare and contributing institution
Controlled researchQualified users under governed access
Optional public releaseDe-identified data chosen for public use
One consent. Four intake paths.
The contribution route changes. The stewardship contract does not.
Trial-mediated
The trial adopts a common addendum.
- The protocol would adopt a Clarida consent addendum written for IRB review.
- The sponsor or site signs a data contribution agreement.
- Clarida would receive the data and run the standard ingest contract.
- De-identification, tiering, hosting and access are handled from then on under the same rules.
Institution-mediated
One contribution agreement replaces pairwise transfers.
- The institution contributes under a reusable contribution agreement.
- Each dataset carries consent and permitted-use metadata.
- A time-limited embargo protects planned contributor publications.
- At release, approved users discover and analyze the same harmonized objects.
Patient-mediated
The patient authorizes retrieval. Clarida would move the data.
- The patient signs the portable consent and a retrieval authorization.
- On that authorization, Clarida requests records from hospitals, laboratories and sequencing vendors.
- The patient receives a clear data inventory and research summary.
- The patient may then select controlled sharing or optional public release.
Clinical findings would be returned only through an approved clinical process. Nothing on this page is medical advice.
Existing cohorts
A cohort joins under the consent it already has.
- The original consent language is mapped to DUO terms and profile fields. The narrower restriction always wins.
- The IRB of record makes a determination on secondary use and on reliance.
- Data enter only the tiers the original consent supports. Nothing is widened by default.
- A re-consent invitation goes only to living participants. Choices already recorded persist for participants who have died, and a family may still ask for prospective withdrawal.
What a trial team would no longer build alone
Stewardship becomes shared infrastructure.
- IRB-ready consent and authorization language
- Data-use and contribution agreement templates
- De-identification for structured data, free text and DICOM
- Aggregate, controlled, secure-workspace and federated access modes
- Independent data access committee operations
- One access request across everything the Foundation holds
- Contributor attribution and publication embargo controls
- A withheld evaluation tier for benchmark studies
- A dataset DOI and citation for every contribution
- Your own data returned harmonized, with treatment and outcome correlates attached
Open question. Federally funded genomic data must still be deposited in a designated federal repository under the applicable genomic data sharing policy, and the Foundation cannot replace those obligations. The design needs an approved-repository or trusted-partner arrangement so those datasets can be reached through the same request.
Enforced at query time
Consent is executable metadata.
FHIR ConsentStores the patient's current choices and consent history.
DUO + profileEncodes allowed purposes, restrictions and permissions not covered by DUO.
GA4GH PassportsCarries verified researcher identity, affiliation and access claims.
Policy checkTier, purpose, embargo, jurisdiction and consent version are checked before a query runs.
Access modeDownload, secure workspace or federated compute. The jurisdiction tag decides which modes are allowed for each object, so data that cannot leave an enclave can still be analyzed in place.
Audit logRecords who accessed what, why, when and under which consent version.
Withheld evaluation tier
A held-out set, so the field can learn whether the models work.
Multimodal models for rare pediatric cancers cannot be judged on data they were trained on. The design keeps a sequestered evaluation set for exactly that purpose, and patients and contributors would see it as a named use rather than a side effect.
Who opts inPatients, through the separate benchmark permission. Contributing sites, through the contribution agreement.
Who sequestersThe data access committee assigns records to the held-out set. An independent benchmark host administers evaluation. Contributing labs never hold the key.
How scoring worksModels run against the held-out set inside the enclave. Results report metrics, never records, and the set itself is never released.
How it stays usefulThe set refreshes as new contributions arrive, so the benchmark tracks the disease rather than one snapshot.
Pediatric and lifecycle handling
Consent changes as the patient grows.
A parent or legally authorized representative gives permission. The child gives assent when developmentally appropriate. Re-consent would be requested when the participant reaches the legal age of adulthood. The patient would be able to review scopes, change future permissions, see access history and request withdrawal at any time.
Withdrawal is prospective. The Foundation would stop new use and future release when the withdrawal takes effect, subject to legal and research-record duties. Data already released publicly, included in completed analyses or published cannot be recalled. If a participant dies, the choices on record continue to govern the data, and a parent or legal representative may still request prospective withdrawal.
Governance and limits
Plain limits. Independent oversight.
- An independent data access committee will include patient representation.
- The Foundation will act as a neutral steward. It will not sell the dataset or compete with the researchers who use it.
- Transparency reports will describe access requests, approvals, denials, incidents and policy changes.
- Research views may shift dates while preserving clinically meaningful intervals. Date shifting alone does not establish HIPAA de-identification.
- Direct identifiers are removed or protected. Genomic and deeply linked health data can still carry a risk of re-identification.
- Public release is an optional patient decision made with that risk stated plainly.
Legal scaffolding
One operating model. Multiple legal bases.
- US research and privacy law. Broad consent under 45 CFR 46.116(d) with single IRB reliance under 45 CFR 46.114. HIPAA research authorization, de-identification and patient-directed access under 45 CFR 164.508, 164.514 and 164.524.
- Federal data policy. The federal genomic data sharing policy, Certificates of Confidentiality and the information-blocking rules under 45 CFR Part 171.
- International and state law. GDPR Articles 6, 9(2)(j) and 89 with an approved transfer mechanism for European contributors, and state genetic-privacy laws tracked in a jurisdiction matrix.
The full legal basis table, with counsel flags, is in the working specification and is not published here.
Technical and governance precedents: Count Me In; NCI Genomic Data Commons and dbGaP controlled access; GA4GH Data Use Ontology and Data Passports; Sage Bionetworks Elements of Informed Consent; ICGC ARGO consent framework; UK Biobank governance; and Break Through Cancer Data Science Hub. These are precedents, not partners, and not endorsements of this proposed design.
07 / What one real catalog teaches
Heterogeneous inputs are normal
The design must understand incomplete, overlapping and differently processed contributions.
The illustrative catalog includes raw reads, aligned files, reports, spreadsheets, multiple laboratories, FFPE and flash-frozen tissue, shared normals, tumor-only data, report-only deliveries and expected files that have not yet arrived.
Illustrative de-identified source catalog
What comes in, and what would come outMulti-timepoint WES + WTSFFPE · multiple deliveries · raw + processed + reports
Five serial analyses, but one timepoint is report-only and another delivery contains FASTQ without BAM, spreadsheets or checksums.
Reconcile orders and specimens; canonical WES/WTS reprocessing where raw data exist; source-report extraction; explicit raw-data availability and QC fields.
Output: serial DNA/RNA matrices
Output: serial DNA/RNA matrices
Two independent lung nodulesFlash-frozen · WES + WTS · one shared normal
Separate tumor folders use a single germline reference. The relationship is valid but easy to misread without explicit graph metadata.
Confirm fingerprint and normal identity; assign durable lesion A/B IDs; process both tumors against the same normal; enable within-lung heterogeneity analysis.
Output: lesion-resolved comparison
Output: lesion-resolved comparison
PacBio long-read RNAHiFi BAM + Iso-Seq outputs
The folder name references tumor/PBMC WGS and long-read RNA, but only the long-read delivery is present.
Inventory expected versus received modalities; publish the isoform layer; keep tumor and blood WGS as missing/awaited, not absent biology.
Flag: expected WGS missing
Flag: expected WGS missing
Tumor-only DNA + RNAClinical profiling · no matched normal
Paired DNA and RNA reads are available, but inherited variation cannot be filtered with a patient-matched normal from that assay.
Preserve clinical calls; run tumor-only processing with explicit uncertainty; cross-check against the patient's normal from other sources where consent and methods allow.
Flag: tumor-only interpretation
Flag: tumor-only interpretation
Protein-level reportsPhosphoproteome PDF + abundance spreadsheet
Processed results exist without the same raw-data depth as the sequencing folders.
Map protein and phosphosite identifiers; retain original z-score reference and report provenance; label results as report-derived if raw spectra are unavailable.
Flag: report-derived evidence
Flag: report-derived evidence
Planned / pending assaysImmunopeptidomics · single-cell · organoids · phosphoproteomics
Some tests are planned, pending, performed elsewhere or not yet delivered.
Use status values (planned, collected, processing, delivered, failed QC, unavailable) so absence of data is never mistaken for a negative assay result.
Output: complete audit trail
Output: complete audit trail
Missing is not negative.
Discordance is measured, not erased.
Raw and source-native results remain available.
08 / Research-ready by default
The researcher experience
Analysis should begin with a cohort question, not months of reconstruction.
"No prep work" should mean no foundational wrangling. Researchers still choose the scientific question, cohort, endpoint and model. The Foundation would supply a validated common substrate.
01
Immutable source layer
Original FASTQ, BAM, reports, spreadsheets, DICOM and raw instrument files with checksums, source metadata and access controls.
02
Harmonized molecular layer
Canonical alignments, calls, expression, isoforms, proteins, cells and imaging features generated with versioned pipelines.
03
Analysis-ready patient layer
Cross-patient and within-patient tables, common vocabularies, treatment timelines, outcome labels, QC and full provenance.
Example cohort query
Find metastatic sarcoma patients with a recurrent fusion event; compare DNA support, short- and long-read RNA, pathway activity and protein signaling before and after pathway-directed treatment; relate the changes to ctDNA, lesion response and toxicity.
same referencesame IDssame time axisquality filteredprovenance attached
- patient_timeline.parquet
- specimens.parquet
- variants_long.parquet
- copy_number.parquet
- structural_variants.parquet
- expression.parquet
- fusions_isoforms.parquet
- single_cell.h5ad
- proteomics.parquet
- treatment_exposure.parquet
- lesion_response.parquet
- provenance.json
Researchers should never have to spend their first six months reassembling the patient.
09 / Use new tissue to set the standard
When new tissue is collected
Capture each nodule once. Preserve enough material to answer future questions.
The high-value work begins before an operation: coordinate surgery, pathology, the biobank and assay laboratories so each lesion remains distinct and pre-analytic details are recorded at the source.
Map every lesion
Connect preoperative imaging, operative location, specimen label, pathology block and molecular files through one durable lesion ID.
Repeat a core panel
Use comparable DNA, short-read RNA and pathology assays at every major disease state so serial change can be measured directly.
Add deeper modalities
Where tissue quantity and quality allow, add WGS, long-read RNA, single-cell and spatial, protein, immunopeptidome and functional models.
Record the treatment state
Capture exact prior dose, last exposure, drug holds, surgery timing and concurrent therapies, not only "pretreated" or "post-treated."
Each nodule kept separate
FFPE + digital pathology
Diagnosis, tumor content, necrosis, IHC, whole-slide image and future spatial assays.
Flash-frozen tissue
WGS/WES, short-read RNA, long-read RNA, proteome and phosphoproteome.
Viable cryopreserved cells
Single-cell or multiome analysis, organoids and functional drug testing where feasible.
Matched blood / PBMC
Germline reference, sample identity, HLA and immune analysis and normal comparator.
Serial plasma
Pre- and postoperative ctDNA or other blood biomarkers linked to imaging and treatment.
Pre-analytic metadata
Collection time, cold ischemia, preservation, aliquot, storage and transfer conditions.
Clinical diagnosis and required pathology always take priority. Research allocation must follow a pre-approved collection plan.
The data-readiness commitment
The research begins where data wrangling ends.
An authorized researcher should be able to open a notebook and analyze across patients, timepoints, modalities and treatments without first reconstructing what each file means.
- 01Every source file is immutable and checksum-verified.
- 02Every specimen and derived result has traceable lineage.
- 03Every harmonized output is versioned and reproducible.
- 04Every missing or uncertain element is explicit.
- 05Every new timepoint refreshes the longitudinal record.
- 06Every record carries the patient's consent, machine-readable and enforced at query time.
Clean once. Align once. Validate once.
Analyze repeatedly, by patients, researchers, drug developers and future care teams.
- NCI GDC-inspired graph + harmonizationCase, biospecimen, assay and file lineage; validation; common processing.
- GA4GHVRS, Phenopackets, data-use terms and portable, federated workflow interfaces.
- HL7 FHIR + OMOPComputable genomic reporting and standardized longitudinal clinical data.
- CDISC + DICOMRegulatory trial datasets and interoperable imaging with structured lesion tracking.
- CWL / containersPortable, reproducible bioinformatics workflows with explicit inputs and outputs.
- BioCompute provenanceHuman- and machine-readable workflow versions, parameters and execution history.
- Controlled vocabulariesGene, variant, disease, drug, lab, toxicity, anatomy and outcome concepts mapped once.
- Governed accessPublic aggregate knowledge; controlled individual-level data; secure computation when needed.
- DUO + FHIR ConsentMachine-readable permitted-use terms attached to a versioned patient consent resource.
- GA4GH PassportsVerified researcher identity and access claims used by the policy check.
- Single IRB relianceA master addendum and reliance model for cooperative research where applicable.
- Common Rule + HIPAABroad consent, authorization, access, de-identification and withdrawal rules.
- GDPR + transfer controlsResearch basis, safeguards, data-subject rights and an approved international transfer mechanism.
Reference architecture informed by the NCI Genomic Data Commons data model, data processing and quality framework; GA4GH Phenopackets, Variation Representation Specification and Data Use Ontology; HL7 FHIR Genomics Reporting; OHDSI OMOP; CDISC SDTM; Common Workflow Language; and NCI Biospecimen Best Practices. These are precedents, not partners. This design is not a final implementation specification and will be co-designed with pathology, bioinformatics, privacy, regulatory, clinical and patient experts.