DNAnexus
DNAnexus runs a cloud platform where diagnostic laboratories, biopharma companies, health systems and biobanks store, process and analyze genomic and clinical data. Titan runs sequencing workflows at production scale and tracks their versions and execution history. Apollo joins molecular data to clinical and phenotypic records so researchers can build cohorts, study biomarkers and test trial feasibility. Two agents sit on top. Omics Data Agent turns a plain language description of a cohort into a query against the dataset's data dictionary, places the cohort in Apollo's Cohort Browser and shows the SQL behind it. AutoML Agent proposes analyses and preliminary models, such as whether patient characteristics predict treatment response, and returns an execution plan and a notebook the researcher can check, with MLflow and Ray packaged in.
The Mountain View company was founded in 2009 and is led by CEO Thomas Laur. It counts more than 64,000 active users, 140 petabytes of multimodal data and 3 billion API calls a month. Customers include Natera, Regeneron Genetics Center, Lilly, Merck, GSK, Novartis, St. Jude and Our Future Health, and the platform runs precisionFDA for the FDA. UK Biobank built its Research Analysis Platform on DNAnexus, where researcher access is currently suspended. CariGenetics launched a Caribbean genomics research environment on the platform with NVIDIA in October 2026. Licenses are set by order form with usage billed monthly, and HIPAA projects run under a separate business associate agreement.
Capability Axes
An AI Health Index grade measures what a buyer can verify from public sources on the date shown. It is not a rating of how good the product is. A vendor can build an excellent system and grade low on an axis because it publishes nothing an outsider can check. How grades read
Two native agents extend the platform with functions built around models. Omics Data Agent interprets a researcher's request against a data dictionary and builds a cohort in Apollo, and it can also summarize data and run longitudinal queries. AutoML Agent turns questions about a cohort into exploratory analyses, preliminary predictive models and survival analyses, packaged with MLflow and Ray.
The rest of the platform stands without them: Titan runs and tracks sequencing pipelines at a rate the company puts above 500 million reads an hour, Apollo offers a conventional Cohort Browser, and the Trusted Research Environment governs shared access to sensitive data. The conversational cohort builder and the model development copilot are the parts that would not exist without their models.
Both agents show their work before anything is used. Omics Data Agent can run a query and create a cohort in Apollo, and it exposes the SQL so analysts can inspect the filters and refine the result. For complex analyses, AutoML Agent presents an execution plan for review first and then generates a notebook that records each step.
Its usage terms require a qualified researcher to validate outputs before they are used in consequential work, and platform permissions set which projects and data each user can reach. The review route is concrete for both agents. What is not stated is when an agent should stop and hand back, for instance on a request it cannot map to the data dictionary or a model trained on too few cases.
The cohort agent maps a question to the dataset's schema and data dictionary, then generates SQL against it. AutoML Agent uses a language model to propose analyses and write code, and its notebooks carry model evaluation, feature attribution and reusable steps, so researchers can inspect the models built for their own projects across several modeling frameworks. That makes each output traceable.
What produces the conversational layer is not named: no model family or provider is identified for either agent, and nothing describes their training data or how agent releases are versioned. DNAnexus has added NVIDIA NIM and CUDA-X microservices to the platform for drug discovery workloads, which describes the toolchain around the agents rather than the model inside them.
Projects run on Amazon Web Services, Microsoft Azure or Oracle Cloud Infrastructure in the region the customer picks, and the platform keeps a project's files and jobs in that region. AutoML conversation history stays inside the DNAnexus project under the project's access controls, and a researcher shares results only with authorized members. The master agreement treats external platforms and customer supplied tools as separate from the hosted subscription, each under its own terms.
The model layer is the gap. Neither agent names the language model provider that receives a prompt, whether inference runs inside the customer's cloud region, or what retention and training terms apply once a query reaches the model.
The platform's record rests on scale and named customers. UK Biobank built its Research Analysis Platform with DNAnexus, Natera chose it to run genetic testing across distributed sites, and Regeneron Genetics Center used an earlier Titan configuration for UK Biobank exome processing at about four hours a sample. The company reports 90% customer satisfaction (July 2026), more than 64,000 active users and 50 billion files and records.
For the agents, the one figure is the company's estimate that AutoML can speed model prototyping by up to 80%, with no named deployment, design or baseline behind it. Platform results predate the agents and do not carry to them, and researcher access to the UK Biobank environment is currently suspended.
Customer data serves the customer alone unless a separate data use addendum allows more. AutoML conversation data stays in the project, the AutoML privacy terms rule out using it to train or fine tune models, and a researcher can delete the conversation, its logs and other artifacts from project storage. Projects holding protected health information need a signed business associate agreement and the designated HIPAA project settings.
After termination the master terms allow account data to be deleted, though no fixed deletion window is set and Omics Data Agent has no retention statement of its own. The AutoML usage policy assigns scientific validation to the researcher, and no procedure is set out for an agent safety incident such as a wrong cohort reaching an analysis.
The business associate agreement can be read before contracting and names DNAnexus as the business associate. Its scope is limited to projects explicitly designated as HIPAA Projects, which the covered entity marks in the user interface and configures with the required safeguards, encryption included.
The subscription agreement bars protected health information from the platform until a signed agreement is in place, and the agreement binds subcontractors that handle protected health information to the same restrictions and safeguards. Any project outside the designation sits outside the agreement.
A laboratory running research and clinical work side by side keeps that boundary in its project setup, and agent work inside a designated project falls under the same terms as the rest of the project.
Schellman certifies the DNAnexus information security management system under ISO/IEC 27001:2022, on public certificate 1623622-4, with a scope covering the design, development, deployment and operation of the genomic data platform. The trust center at trust.dnanexus.com presents the certificate, the statement of applicability and a redacted 2026 penetration test summary.
The federal marketplace lists the platform under the FedRAMP agency path at Class C, the Moderate designation, and the platform hosts the FDA's precisionFDA environment along with GxP workloads through its own GxP solution. The certificate covers the platform as a whole, so the agents fall inside its scope as platform features, while project contracts govern the tools and data in each deployment.
Neither agent is sold for clinical use. AutoML's intended use is initial data exploration and model prototyping, and its terms prohibit using it for clinical diagnosis or to guide treatment. Omics Data Agent builds research cohorts, with the SQL open to inspection. Titan supports laboratory pipelines that customers validate themselves under their own laboratory regulations, and the GxP solution supports regulated work for sponsors.
The FDA is a customer, running precisionFDA on the platform, which is distinct from any device authorization. The planned Oracle Health integration, which would bring genomic results to the point of care, is the one route toward clinical use, and it rests on a nonbinding agreement announced in September 2025 rather than a shipping feature.
AutoML generates an execution plan and an analysis notebook, so researchers can see the variables, features and outputs that went into a model, and Apollo's cohort tools show the composition of the underlying population and how it was selected. By contract, responsibility for the choice and quality of input data sits with the customer. Those tools let a research team examine and correct the analysis it gets back, which is where governance of the agents rests today.
There is no subgroup evaluation of either agent, for instance how cohort queries perform on datasets with different ancestries or coding conventions, and no independent audit of agent behavior across the biobanks and populations the platform serves.
The master agreement gives paid subscribers a limited warranty that the service conforms to its documentation, with correction or a prorated refund as the remedy, inside a claim window and subject to exclusions. The AI terms then place validation on the customer and disclaim the accuracy and completeness of AI outputs. Liability is generally capped at fees paid in the previous 12 months, with a stated $1,000,000 ceiling on the DNAnexus side.
Researchers can trace a cohort or an analysis through the generated SQL or notebook and rerun it. The service remedy covers the platform working as described, not an inaccurate scientific result, so a wrong cohort or a flawed model carries no remedy from the vendor.
Apollo connects clinical and phenotypic records to germline, somatic and expression data in one queryable dataset, and Omics Data Agent writes cohorts back into Apollo, with the SQL reusable in downstream tools. The platform links to laboratory information management, trial management and electronic data capture systems through its API, which handles more than 3 billion calls a month.
On the clinical side, DNAnexus and Oracle Health announced in September 2025 an intention to bring genomic data into select Oracle Health clinical applications, under a nonbinding agreement. No production connection to an EHR is named, for the platform or for the agents, and neither agent reads a chart or writes its output into one.
Each project is assigned a cloud provider and region, and the platform keeps that project's uploaded files and running jobs in the region chosen. AWS, Azure and Oracle Cloud Infrastructure regions are supported, customers can bring their own storage under a separate addendum, and the Trusted Research Environment and Trusted Regulatory Spaces give controlled access to designated data for consortia and regulatory work. Project administrators grant roles and permissions.
That gives research storage and compute a clear location, as at UK Biobank, Our Future Health and the CariGenetics environment in the Caribbean. The agents' own processing route is what stays open: whether a call to the language model leaves the project's region is not stated.
The platform is licensed by order form, with usage billed monthly. The pricing basis separates compute, storage and data transfer, charges vary by region and instance type, and each account administrator sees the customer's negotiated rate card. AutoML access requires its own license.
The one public rate card is UK Biobank's from January 2026, which prices storage outside the supplied dataset at £0.01783 per GB a month and egress at £0.05015 per GB; those rates belong to that environment alone. The standard contract prices the enterprise platform, agent access and professional services separately. Outside UK Biobank, the units of charge are public and the rates come with the order form.
The platform serves diagnostic laboratories, biopharma translational teams, health systems, biobanks and government agencies. A laboratory runs its own sequencing workflows through Titan, as Natera does across its testing sites, while Apollo helps a translational scientist join clinical records to molecular data and compare cohorts; pharma customers include Lilly, Merck, GSK and Novartis.
Biobanks and national programs such as UK Biobank, Our Future Health and CariGenetics host researcher access in the Trusted Research Environment, and the Accelerated Cure Project opened its multiple sclerosis data to outside researchers on the platform in June 2026. Named deployments of the platform back those settings. The two agents are newer and carry no comparable validation in any of them.
Alternatives to DNAnexus
Other products that do the same kind of work as DNAnexus, and what each one does.
Genomics company in San Mateo, California, sequencing with its proprietary Exome+ assay in a CLIA certified, CAP accredited laboratory.
Frekil builds AI agents that run observational real world evidence studies for pharmaceutical, biotech and contract research teams on data they already hold or license, from protocol through final report.
Cloud genomics analysis company (NASDAQ: SOPH) based in Rolle, Switzerland and Boston.
Truveta is owned by 30 US health systems that contribute de-identified health records every day to a shared research dataset.
Rare disease genomic diagnosis company (NASDAQ: WGS) in Gaithersburg, Maryland, running exome and genome sequencing in its own accredited laboratory and interpreting it with automation, machine learning and clinical scientists over GeneDx Infinity, its dataset of prior cases.
MyOme is a genomics company in Menlo Park, California, with a CLIA certified and CAP accredited laboratory, that sells whole genome screening built around integrated risk scores.
Pricing
Vendor-published figures are labeled as such. Figures labeled “Estimated” are derived from third-party sources and have not been confirmed by the vendor.
| Entry Price | Pricing Basis | BAA Tier | Implementation | Source |
|---|---|---|---|---|
|
Custom quote
|
The subscription is set by order form and usage is billed monthly. Compute, storage and data transfer charges depend on region and instance type. AutoML access requires a license. The UK Biobank research rate card applies to that deployment. | A signed business associate agreement covers designated HIPAA Projects with the required security settings. It does not apply to other projects in the account. | Consulting and implementation are purchased through an order form and statement of work. The agreement allows fixed scope pricing or hourly fees, with travel and lodging billed where applicable. | Vendor Published |
The enterprise buyer receives a negotiated rate card. A project administrator can use that card to estimate cost by region and compute type. UK Biobank's January 2026 card lists £0.01783 per GB per month for additional storage and £0.05015 per GB for egress. Those figures apply to the UK Biobank Research Analysis Platform, whose access is currently suspended.