Drug Discovery AI
X

Xaira Therapeutics

South San Francisco company founded in 2023 and launched publicly in April 2024 with more than 1 billion dollars in committed funding, the largest initial funding commitment in ARCH Venture Partners history, co led by ARCH and Foresite Labs with F-Prime, NEA, Sequoia Capital, Lux Capital and Lightspeed among the syndicate. The stated aim is end to end application of AI across drug discovery and development, combining AI methods research, large scale biological and clinical data generation, and therapeutic product development.

Co founders include Marc Tessier-Lavigne, former Chief Scientific Officer of Genentech and former president of Stanford University, who serves as CEO; David Baker, the University of Washington protein design researcher awarded the 2024 Nobel Prize in Chemistry; Hetu Kamisetty, Chief Technology Officer; Vik Bajaj; and Robert Nelsen. The research team includes people who developed the RFdiffusion and RFantibody protein and antibody design models in the Baker laboratory, and the company absorbed technologies and personnel spun out of Illumina's functional genomics R&D effort along with a proteomics group from Interline Therapeutics.

The board includes former FDA Commissioner Scott Gottlieb, Nobel laureate Carolyn Bertozzi and former Johnson & Johnson CEO Alex Gorsky. Headcount was reported at roughly 204 as of May 2026. In March 2026, after two largely silent years, the company made its first substantial platform disclosures: X-Cell, a virtual cell model of roughly 4.9 billion parameters that predicts cellular responses to genetic perturbation, released alongside X-Atlas/Orion, presented as the largest publicly available genome wide Perturb-seq dataset, and a preprint describing the model's development.

The reported training corpus is 25.6 million perturbed single cell transcriptomes across sixteen biologically diverse contexts. The company has since described its therapeutic focus as inflammatory and immunological disease, with antibody therapeutics as the starting point. No clinical programme, named pipeline asset or named pharmaceutical partnership had been publicly disclosed as of this review.

AI Health Index verifiedJuly 28, 2026
Compare Xaira Therapeutics with other vendors
Founded
2023
Headquarters
South San Francisco, CA
Website
xaira.com
Categories
drug-discovery
Indexed Products
X-Cell, X-Atlas/Orion
Assessment

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

AI Capability
AA on AI CentralityThe artificial intelligence is the product. Remove the model and there is nothing left to sell.
Vendor Published

The company exists to apply AI to drug discovery end to end and has no other business, with a stated combination of AI methods research, large scale biological and clinical data generation, and therapeutic product development. Technical provenance is unusually strong: the research team includes people who developed RFdiffusion and RFantibody, among the most influential protein and antibody design models produced in the Baker laboratory, and co founder David Baker was awarded the 2024 Nobel Prize in Chemistry for protein design work.

A qualification belongs on this grade rather than a discount: centrality here describes intent and capability, since no platform or product has been publicly disclosed against which the claim could be tested.

CC on Autonomy and Oversight ModelAutonomy is claimed and oversight is asserted without a mechanism. Human in the loop appears as a phrase rather than a described control.
Vendor Published

Converted from Not Rated. The prior note recorded that there was nothing to assess, which was accurate when written and is no longer.

The company now states publicly that it is building predictive and agentic AI models across the complete spectrum of the drug discovery and development process. That is a direct claim about autonomy, and it is made without any accompanying description of what those models decide, at which points a person reviews or approves, or what follows when an output is wrong.

Its released virtual cell model is described as helping infer which perturbations matter, in which biological contexts, and where experimental follow up is most valuable. That places it in a prioritisation role ahead of laboratory work rather than in an unattended decision role, and laboratory follow up is itself a check on the prediction. But that reading is inferred from a capability description rather than taken from a disclosed oversight design, and the note should not present an inference as a finding.

Graded rather than left unrated because the question was asked and the company has now made a claim that invites it. Asserting agentic capability across the whole discovery and development process while describing no oversight mechanism is a weaker position than saying nothing at all, because the word invites a reader to assume a governance structure sits behind it.

CC on Model and Technology TransparencyThe architecture is described in general terms with nothing identified. Proprietary is asserted rather than explained.
Vendor Published

The provenance is public and the platform is not, and buyers should be careful not to substitute one for the other. RFdiffusion and RFantibody were developed in an academic laboratory, are published and openly available, and represent genuine and checkable prior work by people now at the company. None of that is disclosure about what Xaira has built.

No Xaira specific model, method paper, architecture description, benchmark or technical documentation was located in this review, and the founders' academic publication record should not be read as transparency about the company's own systems. Graded C rather than lower because the technical lineage is real and specific rather than asserted.

DD on Model Supply Chain DisclosureNothing establishes who else sits between a patient record and an answer.
Vendor Published

No data governance, consent or stewardship disclosure was located, no party in the chain is named, and no workflow description exists against which one could be read. Two facts suggest human derived data may fall within scope even though nothing confirms it. The company absorbed technologies and personnel from a large sequencing company's functional genomics effort and a proteomics group from another firm, and it states an intent to generate biological and clinical data at scale.

Clinical data generation at scale is not a chemistry proposition, and a company saying it intends to do that has a stewardship question ahead of it whether or not it has one today. The acquisition point deserves its own line because it recurs across this index: absorbed assets carry absorbed obligations, and a dataset transferred with a team may have been assembled under consents, ethics approvals or institutional agreements that the receiving company did not negotiate and may not have read.

Those conditions do not lapse on transfer. Graded at the floor because nothing is named rather than as a judgement about intent, and the question was asked so the silence is the finding. Ask what consent basis will govern the clinical data the company intends to generate, whether any absorbed assets carry human derived material with conditions attached, and who determined that those conditions permit the current use.

CC on Clinical and Operational EvidenceNamed customers, or vendor reported percentages with no method, denominator or reference standard. Scale of use is recorded here and is not treated as evidence of benefit.
Vendor Published

Essentially none exists, which is a reasonable position for a company that launched publicly in April 2024 but is the fact a buyer needs. No disclosed pipeline, clinical programme, development candidate, named pharmaceutical partnership or published platform result was located in this review.

What is remarkable is the input rather than the output: more than 1 billion dollars in committed funding, the largest initial commitment in ARCH Venture Partners history, a reported valuation around 4 billion dollars, roughly 204 employees, and a founding and board roster of unusual seniority. The index treats capital and credentials as signals about what sophisticated investors believe, not as evidence that a platform works, exactly as it treats deal value elsewhere in this category. Graded C rather than Not Rated because evidence was sought and the absence is itself the finding.

CC on AI Safety and PHI StewardshipGeneral assurances of privacy and security that do not answer the questions artificial intelligence raises: what is retained, what reaches a model, and what happens to it there.
Vendor Published

No data governance, consent or stewardship disclosure was located, and no workflow description exists against which one could be read.

Two facts suggest human derived data may fall within scope even though nothing confirms it. The company absorbed technologies and personnel from Illumina's functional genomics research effort along with a proteomics group from Interline Therapeutics, and it states an intent to generate biological and clinical data at scale. Clinical data generation at scale is not a chemistry proposition, and a company saying it intends to do that has a stewardship question ahead of it whether or not it has one today.

Graded C rather than left unrated because the question was asked and the silence is itself the finding, which is the same basis on which this record grades clinical evidence. A buyer should treat this as an open item to revisit rather than a settled position, and should ask what consent basis governs the data the company intends to generate, and whether any of the absorbed assets carry human derived material with conditions attached.

Regulatory and Compliance
BB on HIPAA and BAA PostureBusiness associate status is stated and supported by a substantive privacy document, with the agreement or its scope not fully published. For a vendor outside the United States, an equivalent regime documented to this depth grades here.
Vendor Published

Not applicable on the current disclosure. The company is a research organization rather than a business associate of covered entities, and no relationship involving protected health information was identified.

CC on Security Certifications and Trust CenterControls are described with an outside check behind them, such as independent penetration testing on a stated cadence, but no attestation against a recognised framework.
Vendor Published

Converted from Not Rated. No SOC 2, ISO 27001 or equivalent attestation was located and no trust centre was found. The company is private, so the annual report route that supplies a cybersecurity governance disclosure for the listed vendors in this category does not exist here.

Retained from the prior assessment and still correct: no partner relationship has been disclosed that would currently place a third party's proprietary targets or data inside this environment, so the absence carries less weight here than for vendors already holding partner material. The company has said that partnerships will be central to its next phase, so that mitigation is temporary and this row should be re examined once any collaboration is announced.

What has changed since the prior assessment is the direction in which data moves. The company now publishes outward: a genome wide perturbation dataset presented as the largest publicly available one of its kind, a released model, a preprint, and a stated intention to make further elements of the model and its underlying data available to other scientists. An outward release programme carries control questions that an attestation does not answer and that this axis should record. What is reviewed before a dataset or a model is published, who approves a release, whether any use restrictions attach to the released artefacts, and whether anything in a perturbation corpus could be traced back to the individuals whose cells were the source. Nothing describing that process was located.

CC on FDA and Regulatory StatusNo device claim is made and the product is scoped accordingly. Most administrative and operational products sit here and are not penalised for it, because this axis grades the appropriateness of the positioning rather than possession of a clearance.
Regulatory Filing

No regulatory record exists. No IND, registered clinical trial, development candidate or disclosed regulatory interaction was located in this review, and no product or platform is presented as a regulated device. Given the launch date this is a statement of stage rather than of conduct, but a buyer comparing platforms in this category should be clear that the largest initial funding commitment in the category has not yet produced a publicly disclosed regulatory milestone.

BB on AI Governance and Bias DisclosureA governance framework with named process behind it, such as certification to an artificial intelligence management standard, or material written for a customer own review committee to evaluate the product with.
Vendor Published

Upgraded from Not Rated. This row was written when the company had disclosed almost nothing, and that is no longer the position.

In March 2026 the company released X-Cell, a virtual cell model of roughly 4.9 billion parameters that predicts how cells respond to genetic perturbation, and described its development in a publicly posted preprint. More significant for this axis, it also released X-Atlas/Orion, presented as the largest publicly available genome wide Perturb-seq dataset, with the model's training corpus reported at 25.6 million perturbed single cell transcriptomes across sixteen biologically diverse contexts. The company has stated it intends to make elements of the model and its underlying data available to other scientists.

Releasing the training corpus publicly is the strongest available answer to the representativeness question in this domain, and it is better than describing the corpus, because an outside party can characterise the composition independently rather than accepting the vendor's own account of its diversity. This index credits the same behaviour elsewhere in the category, where a company released the dataset on which its own paradigm can be tested by others.

Held at B rather than A on the basis applied consistently across this lane. There is no formal governance framework, no model card and no statement of intended and unsuitable use. The preprint is not peer reviewed. And the diversity claim is about cellular context rather than about the genetic backgrounds of the donors whose cells underlie perturbation screens, which is the representativeness question this index has pressed on comparable platforms and which remains unanswered here.

One further matter is recorded as company context, because a buyer assessing a research organisation will encounter it. The chief executive resigned as president of Stanford University in July 2023 following a review of papers he co authored earlier in his career. The review panel concluded that he did not personally engage in research misconduct in any of the twelve papers examined, while finding serious flaws in the presentation of data in five papers where he was principal author, apparent manipulation by others in at least four of those, and shortcomings in laboratory oversight and management. This concerns work at prior institutions and no allegation concerning this company was located.

CC on AI Liability and RecourseMechanisms exist that let someone challenge an output, such as audit trails, source traceability or review before commit, with nothing standing behind the output and no route for the harmed party.
Peer Reviewed Publication

The provenance is public and the platform is not, and a buyer should be careful not to substitute one for the other. The protein design and antibody design methods associated with this team were developed in an academic laboratory, are published and openly available, and represent genuine and checkable prior work by people now at the company, which is a stronger form of pedigree than a list of prior employers because the artefacts can be run and criticised by anyone.

None of that is disclosure about what this company has built. No company specific model, method paper, architecture description, benchmark or technical documentation was located in this review, and no warranty, indemnity or remediation commitment attaches.

The distinction is the one this backfill has had to draw repeatedly and it belongs on the final record as clearly as on any: a founder's academic publication record is evidence about people, not about a product, and a reader assembling confidence from papers the company did not publish about systems it does not describe is reasoning about a different object.

Graded above the floor because the lineage is real and specific rather than asserted, and because open methods at least establish what the starting point was. Ask what differs between the published academic methods and the deployed platform, what has been built since, and for a prospective success rate on partner programmes with a denominator.

Integration and Deployment
BB on EHR and Interoperability DepthNamed systems with read access or one directional writing, or standards support with named deployments behind it.
Vendor Published

Not applicable. This is a discovery organization with no provider workflow surface and no EHR touchpoint.

CC on Deployment Model and Data ResidencyA single hosted option with location implied rather than committed.
Vendor Published

No product, platform access model, licensing arrangement or partnership structure has been publicly disclosed, so there is nothing deployed and no counterparty for whom tenancy or residency terms would apply.

That is not the same as an axis that does not reach the business. Every peer in this category has a commercial surface of some kind, whether licensed software, a partnership structure or an open release, and each of those carries its own answer to where data sits. This company has disclosed none, which means a prospective partner cannot establish, before entering discussions, whether its own target information would be processed in the company's environment, in a shared one, or alongside data from others.

Graded C rather than left unrated because the question was asked and the silence is itself the finding, which is the same basis on which this record grades clinical evidence and stewardship. Revisit when the company discloses a commercial model, at which point this becomes assessable on ordinary terms.

Commercial
CC on Commercial TransparencyNo price is published and the posture is discoverable: a buyer can establish how the product is sold and what drives the cost before contacting the vendor. Most of the index sits here.
Vendor Published

Capital transparency without commercial transparency. The funding position is public and specific, at more than 1 billion dollars committed at launch, co led by ARCH Venture Partners and Foresite Labs with F-Prime, NEA, Sequoia Capital, Lux Capital and Lightspeed among the syndicate, described as the largest initial funding commitment in ARCH's history, with a reported valuation around 4 billion dollars. Governance and leadership rosters are equally public.

What is absent is everything a counterparty would need: no disclosed business model, no partnership terms, no revenue, no stated route by which an external organization could access the technology. A buyer cannot currently determine what, if anything, is for sale.

CC on Setting and Specialty CoverageCoverage is claimed broadly without specifics, or stated clearly with nothing validating it yet.
Vendor Published

Breadth is asserted at the maximum and demonstrated nowhere, which is the weakest form of coverage claim. The stated ambition is end to end application of AI across the whole of drug discovery and development, and the assembled capabilities, spanning protein and antibody design provenance, functional genomics from the Illumina spin out and proteomics from Interline, are consistent with that scope. No therapeutic area focus, modality commitment or disclosed programme was located against which to assess it. Buyers should treat coverage as undefined rather than universal.

Comparisons

Compared With

Each comparison carries a written verdict, the buyer conditions that favor each vendor, and a graded side by side. Pairs that cross a category boundary are grouped separately, and their verdicts state where the boundary sits rather than manufacturing a head to head.

Commercial

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
Not applicable
No commercial model disclosed. No partnership, licensing or service structure was identified in public materials. Not applicable. Vendor Published

A buyer cannot currently determine what is for sale. No business model, partnership terms, licensing arrangement, revenue or route of external access has been publicly disclosed, and no named pharmaceutical collaboration was located in this review.

What is public is the capital: more than 1 billion dollars committed at the April 2024 launch, described as the largest initial funding commitment in ARCH Venture Partners history, co led by ARCH and Foresite Labs with F-Prime, NEA, Sequoia Capital, Lux Capital and Lightspeed participating, at a reported valuation around 4 billion dollars.

Governance is equally visible, with a board including former FDA Commissioner Scott Gottlieb, Nobel laureate Carolyn Bertozzi and former Johnson & Johnson chief executive Alex Gorsky. This is capital transparency without commercial transparency, and the distinction matters: knowing who funded a company is not knowing how to work with it.