# Top 5 Controlled Access Repositories for Genomic and Clinical Data An open repository lets anyone download a dataset, while a controlled access repository releases individual-level data only after an approved application that names the requester, the project, and the permitted uses. A workable rule of thumb for choosing: apply where the data were collected under consent conditions that match your research question, and where you can realistically satisfy the gatekeeper, whether that gatekeeper is a committee, a data custodian, or the participants themselves. This controlled access [genomic data repository](/knowledge/bioinformatics/genomic-data-repositories-navigating-public-databases-for-research) comparison covers the five options that matter most for genomic and clinical data. dbGaP anchors human genotype-phenotype research, the Genomic Data Commons anchors cancer research, the Aotearoa Genomic Data Repository governs species data under indigenous guardianship, custodian-controlled linkage repositories manage linked health records, and decentralized platforms hand control to individual data owners. For each option you will see how access works, the pros and cons, and the projects it fits best. ## What Controlled Access Means Genomic data is identifying data. A file of genome-wide variants linked to a diagnosis can sometimes be traced back to a person, which places health information at the sensitive end of the data sharing spectrum [1]. The genomics community has responded with three access models. Open access publishes data with no gate. Registered access lets credentialed users self-certify their eligibility and download without a per-study review. Controlled access requires an approved application for each use. For individual-level human genomic and clinical data, controlled access remains the standard model. Controlled access balances two goals that pull against each other: broad reuse of expensive public data, and respect for the privacy and consent of the participants who contributed it. Funders endorse research data sharing through controlled channels because it optimizes the use of public funds and increases the [statistical power](/blog/guides/statistical-power-what-it-is-and-why-it-matters-in-research) of databases [2]. In practice, most repositories delegate decisions to a data access committee (DAC), the body responsible for reviewing and managing requests [2]. DAC review is not automatically neutral. Governance analyses flag conflicts of interest when data producers sit on the committees that control their own datasets, and they urge that possible data withholding by producers receive thorough attention when access procedures are structured [3]. A global survey of DAC members later found that operations and systemic issues are common across repositories worldwide, with members working to balance review efficiency, quality, and compliance [4]. Controlled access is a governance choice rather than a law of nature. In an open access pilot from the Texas Cancer Research Biobank, many cancer participants were willing to share their genomic data freely, and matched tumor-normal sequences from seven cases, produced by whole exome and [whole genome sequencing](/blog/guides/whole-genome-sequencing), were released without a data use agreement, with a ban on re-identification attempts as the only restriction [5]. Whole [exome sequencing](/blog/guides/exome-sequencing) reads the protein-coding portion of the genome, while whole genome sequencing reads all of it. Between the extremes, the sharing spectrum now also includes public aggregate datasets, federated analysis where code travels to the data rather than data moving to the analyst, and synthetic data that mimic real records [1]. ## Controlled Access Genomic Data Repository Comparison at a Glance The table condenses the comparison. The sections that follow explain each row on its own terms, and the pros and cons section turns the differences into decisions. | Repository | Steward | Data focus | Who approves access | Distinctive feature | Best fit | |--|--|--|--|--|--| | dbGaP | National Institutes of Health, hosted at NCBI | Individual-level human genotype, sequence, and phenotype data | Study-level data access committee | View-only Data Browser for general research use studies [6] | Human genotype-phenotype questions | | Genomic Data Commons (TCGA) | National Cancer Institute | Cancer multi-omic and clinical data | Open tier for processed data, controlled tier for sensitive files | Harmonized data with documented concordance to the legacy build [7] | Cancer biomarker and prognostic modeling | | Aotearoa Genomic Data Repository | A within-nation New Zealand initiative | Non-human genomic data from taonga species and environmental samples | Kaitiakitanga holders (iwi, hapū, whānau) | Māori Data Sovereignty governance [8] | Samples originating in Aotearoa New Zealand | | Custodian-Controlled Data Repository | SA NT DataLink, Australia | De-identified linked health and administrative records | Data custodians | Secure remote analysis in SURE with two-factor authentication [9], [10] | Population health linkage projects | | Nebula and decentralized platforms | Individual data owners | Personal genomic data | The data owner sets the policy | Equitable compensation plus cryptographic access control [11], [12], [13] | User-held sharing and privacy-preserving analysis | ## The Five Repositories in Detail ### 1. dbGaP: The Human Genotype-Phenotype Archive ![Four step flow from submitting a data access request through committee review and download to local analysis.](/article-images/illustrations/controlled-access-genomic-data-repository-comparison-dbgap-request-path.webp) dbGaP, the database of Genotypes and Phenotypes, is the National Institutes of Health archive for studies that link individual-level genotype and sequence data to phenotypic features, maintained at the National Center for Biotechnology Information [6]. Its controlled-access tier holds the individual-level records, and a full request historically followed a fixed path: submit a data access request, wait for the study's data access committee to review it, download each dataset, and examine it locally [6]. Two features make dbGaP easier to work with than that path suggests. The dbGaP Data Browser provides view-only access to summary-level information and to individual-level genotype and sequence data from general research use (GRU) studies, a permission class that permits broad research without disease-specific restrictions [6]. The browser was built in response to requests from the scientific community, and it fills a gap because open genomic resources such as the Exome Variant Server and ExAC publish only summary statistics and aggregate allele frequencies, never individual-level records [6]. The National Institutes of Health expects the browser to decrease unnecessary download requests while still supporting responsible genomic data sharing [6]. ### 2. The Genomic Data Commons and The Cancer Genome Atlas The Genomic Data Commons (GDC), the National Cancer Institute archive that houses The Cancer Genome Atlas (TCGA), is the reference resource for cancer genomic and clinical data. TCGA spans five molecular platforms: mRNA expression, miRNA expression, single nucleotide variants (SNVs), [DNA methylation](/blog/guides/dna-methylation), and copy number alterations (CNAs) [7]. SNVs are single-letter changes in the DNA sequence. CNAs are gains or losses of large chromosome segments, and both are hallmarks of tumor genomes. Legacy TCGA data were aligned to the GRCh37 reference genome, also called hg19. The GDC harmonized them to the current GRCh38 (hg38) build using updated software, pipelines, and annotations [7]. A systematic comparison of the two versions found them very highly concordant, and the published analysis supplies gene lists and mitigation strategies for the differences that remained, so research teams can choose either version with open eyes [7]. The GDC follows the tiered pattern common to large cancer resources, with openly downloadable processed data alongside a controlled tier for the files that carry the most re-identification risk. ### 3. The Aotearoa Genomic Data Repository (AGDR) The Aotearoa Genomic Data Repository is a secure within-nation repository for non-human genomic data from biological and environmental samples originating in Aotearoa New Zealand [8]. Its defining feature is governance. The repository follows the principles of Māori Data Sovereignty and enables kaitiakitanga, the right of guardianship, so that iwi, hapū, and whānau, meaning tribes, kinship groups, and families, can exercise their responsibilities as guardians over biological entities they regard as taonga, meaning precious or treasured [8]. Access decisions rest entirely with the holders of kaitiakitanga over each dataset. No data are released to a requesting researcher until the request has been approved and the access conditions, which can vary by dataset, have been agreed [8]. The repository is nonetheless designed to make data findable by researchers and interoperable with data held in other genomic repositories, so culturally grounded control does not come at the cost of discoverability [8]. ### 4. The Custodian-Controlled Data Repository (CDDR) Custodian-controlled repositories solve a different problem: clinical and administrative records that must be linked across organizations. SA NT DataLink, a data linkage service in Australia, established the Custodian-Controlled Data Repository, a highly secure virtual repository for data storage, analysis, and access [9]. Data custodians, the organizations that own the source records, remain the principal curators, while a dedicated Data Integration Unit performs quality control, integrates datasets, and delivers them for approved linkage projects under custodian oversight [9]. Long-term storage in the repository reduces custodian workload by cutting the frequency of content data provision down to periodic updates, and feedback loops built into the quality control process let custodians improve their datasets by fixing the issues the unit discovers [9]. Researchers reach the data through the Secure Unified Research Environment (SURE), a secure remote access data laboratory operated by the Sax Institute in Sydney, using thin clients and two-factor authentication, so analyses run inside the secure environment instead of on a personal laptop [10]. Participation is opt-in for custodians, and the data remain under their control throughout [10]. ### 5. Decentralized Platforms: Nebula, Derepo, and Cryptographic Access The fifth option is less a single archive than an architectural shift. Nebula is a decentralized platform for genomic data generation, sharing, and analysis that vests control over data in individual owners, ensures transparency and privacy, facilitates regulatory compliance, minimizes expensive data transfers, and shifts sequencing costs from consumers, patients, and biobanks to researchers in industry and academia through an equitable compensation model [13]. The economic logic is straightforward: sequencing a genome now costs under $1,000, down from more than $3 billion for the first [human genome](/blog/guides/human-genome), so the bottleneck is no longer data generation but governance and sharing. Supporting technologies show what decentralized control looks like in practice. Derepo is a privacy-preserving repository that pairs decentralized access control built on distributed ledger technology, which adds trustworthy properties such as Byzantine fault tolerance, the ability to keep functioning when some participants fail or misbehave, with fully homomorphic encryption, a form of encryption that lets authorized parties compute on encrypted data without ever seeing the underlying values [12]. On the access control side, the P2GT scheme uses key-policy attribute-based encryption, a cryptographic method where data are encrypted against a policy of attributes and only matching keys can decrypt, to enforce fine-grained genomic data access and to run genetic tests by matching digitized single nucleotide polymorphisms (SNPs), common single-letter variations, directly on encrypted data, with practical performance demonstrated on the 1000 Genomes dataset [11]. ## Pros and Cons of Each Repository Dataset size matters less than governance, because governance determines what you can do after approval. Here is the pros and cons summary for each option. ### dbGaP Pros: - The deepest public collection of individual-level human genotype and phenotype data - View-only browsing through the Data Browser lets you scope a study before committing to a request [6] - Long-standing committee review with defined permission classes such as GRU [6] Cons: - One request per study, each subject to committee review and its wait times - Consent groups limit reuse, so a disease-specific dataset cannot serve every question - The browser is view-only, so computation still requires an approved download [6] ### Genomic Data Commons (TCGA) Pros: - Multi-omic depth that supports integrative modeling - Harmonized hg38 data with documented concordance to the legacy build [7] - Clinical characterization paired with molecular data in one archive Cons: - Reference build choice affects interpretation, and mixing builds demands the published mitigation strategies [7] - The most identifiable files still sit behind the controlled tier ### Aotearoa Genomic Data Repository Pros: - Governance aligned with Māori Data Sovereignty and kaitiakitanga [8] - Findable and interoperable despite controlled release [8] Cons: - Access conditions vary by dataset and require agreement with the guardians [8] - Scope covers non-human data from samples originating in Aotearoa New Zealand [8] ### Custodian-Controlled Data Repository Pros: - Custodians keep control of their records in an opt-in design [10] - Quality control feedback loops improve the source datasets over time [9] - Secure remote analysis with two-factor authentication keeps data in place [10] Cons: - Analysis happens inside the secure environment, not on your own hardware [10] - Availability depends on custodian participation and capacity ### Decentralized Platforms Pros: - Control rests with individual data owners, with compensation for sharing - Cryptographic designs enforce access and even permit tests on encrypted data [11], Cons: - Early-stage governance with a shorter institutional track record than established archives - Encryption and ledger technologies add technical complexity for research teams ## How to Evaluate a Repository Before You Apply Any data repository comparison should look past the dataset list to the institution behind it. A recent data stewardship framework analyzed repository guidance from the Research Data Alliance, the Confederation of Open Access Repositories, Science Europe, FAIRsharing, and DSpace, and distilled 110 evaluation criteria split evenly between mandatory and optional elements [15]. The framework emphasizes repository-level stewardship, including governance, preservation, provenance tracking, integrity safeguards, and access infrastructure, rather than the scientific validity of any individual dataset [15]. Translate those criteria into questions you can ask before applying. Who decides on access, and are they independent of the data producers, given the documented risk of producer conflicts of interest [3]? How do requests flow, and where do committees report bottlenecks [4]? Does the repository track provenance and commit to preservation over time? Ethical analysis reaches a matching conclusion from another direction: durable progress in health data sharing depends less on unrestricted openness than on calibrated access, privacy-preserving architectures, clear accountability, and sustained public engagement [1]. ## Worked Scenarios **Scenario 1: a prognostic model that combines data types.** You want to predict outcomes in breast cancer using expression, microRNA, mutation, and copy number data together. TCGA through the GDC is the fit. Benchmarking across the breast, ovarian, lung adenocarcinoma, and glioblastoma cohorts showed that integrating the four [genomic data types](/knowledge/bioinformatics/genomic-data-types-a-primer-for-bioinformatics-beginners) produced prognostic models with slightly higher performance than any single data type [14]. Download the harmonized hg38 versions, and if you must combine them with legacy hg19 data, apply the published concordance findings and mitigation strategies first [7]. **Scenario 2: a genotype-phenotype association in humans.** You want to test whether variants associate with a measured trait. Browse before you request. The dbGaP Data Browser exists precisely so researchers can check relevance through view-only access before filing a full application [6]. **Scenario 3: sequencing a taonga species.** Your team sequences a species sampled in Aotearoa New Zealand that local iwi regard as taonga. AGDR is the within-nation option built for this case, with access decisions made by the kaitiakitanga holders rather than by a distant committee [8]. **Scenario 4: linked population health records.** Your project joins hospital admissions with primary care and administrative records. A custodian-controlled linkage repository such as the CDDR keeps custodians in charge of their records while a dedicated unit handles quality control and delivery [9], and you analyze inside SURE rather than downloading copies [10]. ## Common Mistakes and Limitations ![Decision path asking who holds authority over a dataset and routing to a committee, custodian, guardian or data owner.](/article-images/illustrations/controlled-access-genomic-data-repository-comparison-gatekeeper-triage.webp) **Treating controlled access as one thing.** The gatekeepers differ fundamentally: a committee reviews dbGaP requests [2], custodians control the CDDR [10], kaitiakitanga holders govern AGDR datasets [8], and individual owners set policy on decentralized platforms [13]. An application that works for one will not transfer to another. **Assuming review is fast or automatic.** DAC members worldwide report systemic issues and juggle efficiency, quality, and compliance [4]. Build review time into grant timelines from the start. **Reading consent groups loosely.** A GRU dataset permits broad research, but many datasets carry disease-specific restrictions, and using data outside its permission class is a compliance failure rather than a shortcut. **Mixing reference builds casually.** Legacy and harmonized TCGA data differ in predictable ways, and the published analysis provides gene lists and strategies to handle the differences [7]. Check the build before merging datasets. **Assuming no open alternative exists.** Some participants choose open sharing, as the Texas pilot demonstrates [5], and aggregate resources answer frequency questions without any application [6]. **Assuming all repositories meet the same stewardship bar.** Governance, preservation, provenance, and access infrastructure vary, and published criteria now exist for evaluating them [15]. One limitation to keep in view: access decisions are project-specific and depend on your institution, the study's consent terms, and the committee or custodian involved, so treat any comparison as a starting map rather than a guarantee. ## Frequently Asked Questions ### What does controlled access actually restrict? It restricts who can obtain individual-level data and for what purpose. Unlike open downloads, controlled access requires an approved application per dataset, and unlike registered access, it does not rely on self-certification. ### Who reviews and approves a data access request? In most repositories, a data access committee formed for the study or repository reviews requests and manages approvals [2]. Some repositories use other gatekeepers, such as data custodians or community guardians. ### Can I browse controlled-access data without a formal request? For dbGaP, yes. The dbGaP Data Browser provides view-only access to summary-level information and individual-level data from general research use studies, without a full download request [6]. ### Are controlled access repositories only for human data? No. The Aotearoa Genomic Data Repository holds non-human genomic data from taonga species, with access decisions made by indigenous guardians rather than a conventional committee [8]. ### Which repository fits a cancer biomarker project? TCGA at the Genomic Data Commons is the standard starting point, because it pairs multi-omic cancer data with clinical characterization and harmonized reference builds [14], [7]. ### Are decentralized platforms replacing traditional repositories? They complement rather than replace them. Decentralized designs add user-held control, compensation, and cryptographic enforcement, while established archives continue to hold the deepest curated collections, [12], [13]. ## Sources 1. [Ethical Governance of Open Data Across Biomedical Research, Healthcare, and Public Health: Privacy, Equity, Trust, and Controlled Access.](https://pubmed.ncbi.nlm.nih.gov/42752788/) 2. [From the principles of genomic data sharing to the practices of data access committees](https://pubmed.ncbi.nlm.nih.gov/25759363/) 3. [Controlled Access under Review: Improving the Governance of Genomic Data Access](https://pubmed.ncbi.nlm.nih.gov/26720729/) 4. [Achieving Procedural Parity in Managing Access to Genomic and Related Health Data: A Global Survey of Data Access Committee Members](https://pubmed.ncbi.nlm.nih.gov/37192473/) 5. [An open access pilot freely sharing cancer genomic data from participants in Texas](https://pubmed.ncbi.nlm.nih.gov/26882539/) 6. [The dbGaP data browser: a new tool for browsing dbGaP controlled-access genomic data](https://pubmed.ncbi.nlm.nih.gov/27899644/) 7. [Before and After: A Comparison of Legacy and Harmonized TCGA Data at the Genomic Data Commons](https://pubmed.ncbi.nlm.nih.gov/31344359/) 8. [Aotearoa genomic data repository: An āhuru mōwai for taonga species sequencing data](https://pubmed.ncbi.nlm.nih.gov/37712601/) 9. [Supporting the timely, easy and cost effective access to high-quality linked data via the Custodian-Controlled Data Repository](https://doi.org/10.23889/IJPDS.V3I4.690) 10. [Custodian Controlled Data Repository - supporting the timely, easy and cost effective access to linked data](https://doi.org/10.23889/IJPDS.V1I1.346) 11. [P2GT: Fine-Grained Genomic Data Access Control With Privacy-Preserving Testing in Cloud Computing](https://pubmed.ncbi.nlm.nih.gov/33656996/) 12. [Derepo: A Distributed Privacy-Preserving Data Repository with Decentralized Access Control for Smart Health](https://doi.org/10.1109/CSCloud-EdgeCom49738.2020.00015) 13. [Accelerating Genomic Data Generation and Facilitating Genomic Data Access Using Decentralization, Privacy-Preserving Technologies and Equitable Compensation](https://doi.org/10.30953/BHTY.V1.34) 14. [Integration and comparison of different genomic data for outcome prediction in cancer](https://pubmed.ncbi.nlm.nih.gov/26516350/) 15. [Designing Repository-Level Evaluation Criteria: A Data Stewardship Framework for Research Data Repositories](https://doi.org/10.5334/dsj-2026-033)