Xaira Therapeutics
Last reviewed
Jun 7, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 ยท 2,006 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 7, 2026
Sources
18 citations
Review status
Source-backed
Revision
v1 ยท 2,006 words
Add missing citations, update stale details, or suggest a clearer explanation.
Xaira Therapeutics is an American AI-driven drug discovery company based in South San Francisco, California. It launched publicly on April 23, 2024 with more than $1 billion in committed capital, making it one of the largest initial funding commitments in biotechnology history and the largest in the history of its lead backer, ARCH Venture Partners. The company was incubated by ARCH and Foresite Labs and aims to apply foundation-model-scale machine learning and generative protein design to the discovery and development of medicines. Its scientific foundation builds on the computational protein design methods developed in the laboratory of David Baker, the University of Washington biochemist who shared the 2024 Nobel Prize in Chemistry and is a Xaira co-founder. The company is led by chief executive Marc Tessier-Lavigne, a neuroscientist who previously served as chief scientific officer of Genentech and as president of both Rockefeller University and Stanford University. As of 2026 Xaira remains in the research and platform-building stage and has no approved products.
Xaira emerged from roughly six months of stealth operation when it announced its launch on April 23, 2024. The financing of "more than $1 billion" was led by ARCH Venture Partners and Foresite Labs, the company-creation arm of Foresite Capital. ARCH described it as the firm's largest-ever initial funding commitment. A broad syndicate of additional investors participated, including F-Prime, NEA (New Enterprise Associates), Sequoia Capital, Lux Capital, Lightspeed Venture Partners, Menlo Ventures, Two Sigma Ventures, the Parker Institute for Cancer Immunotherapy, Byers Capital, Rsquared, and SV Angel.
Rather than starting from a single academic spinout, Xaira was assembled from several technology and talent sources. It integrated the functional genomics research-and-development unit formerly run inside Illumina, along with a proteomics group and personnel from Interline Therapeutics, and recruited scientists who had built the leading protein and antibody design models in David Baker's lab at the University of Washington Institute for Protein Design. Don Kirkpatrick, a former Interline chief technology officer and Genentech scientist, and Arvind Rajpal, a former Genentech and Bristol Myers Squibb antibody-engineering leader, were among the executives brought in to the founding team.
Marc Tessier-Lavigne was named chief executive at launch. His appointment drew notice because he had resigned the Stanford presidency in 2023 following a review of research conducted in his earlier academic laboratories; the review found that he had not engaged in fraud or personally manipulated data, while flagging problems in papers from labs he led. At Xaira he returned to industry drug discovery, the field in which he had run research at Genentech.
David Baker co-founded the company and serves as a scientific founder and adviser while continuing to direct the Institute for Protein Design. Other co-founders include Hetu Kamisetty, a former Meta AI scientist who became Xaira's chief technology officer in 2025 and leads work on foundational AI models, and Vik Bajaj, the co-founder and chief executive of Foresite Labs and a former Verily and Google X scientist. Debbie Law serves as chief scientific officer.
In early 2025 Xaira expanded its leadership. Paulo Fontoura, who had spent about 16 years at Roche overseeing translational medicine and clinical development across neuroscience, immunology, infectious disease, and rare disease, joined as chief medical officer. Bo Wang leads biomedical AI as a senior vice president. The board of directors includes Robert Nelsen of ARCH, Stephen Knight of Foresite, Vik Bajaj, former FDA commissioner Scott Gottlieb, 2022 Nobel laureate chemist Carolyn Bertozzi, former Johnson & Johnson chairman and chief executive Alex Gorsky, former Genentech research head Richard Scheller, Mathai Mammen, Kaye Foster, and Bryan White.
Xaira's stated goal is to shift drug discovery from an empirical science toward what it calls a more precise engineering discipline by tightly coupling generative AI, large-scale proprietary data generation, and wet-lab biology in an iterative loop. The company organizes its work around three integrated capabilities: foundation-model-scale machine learning, in-house experimental biology, and the generation of very large proprietary datasets to train those models.
On the protein side, Xaira's scientists are associated with the diffusion-based generative tools that came out of the Baker lab, including RFdiffusion and RFantibody for de novo protein and antibody design, and ProteinMPNN for sequence design. These methods extend the structure-prediction advances popularized by tools such as AlphaFold and RoseTTAFold into the generative regime, designing entirely new proteins to hit molecular targets that have resisted conventional approaches. This connects Xaira's work to the broader fields of protein folding and computational protein design.
On the data and cell-modeling side, Xaira has pursued the construction of "virtual cell" models that predict how cells respond to genetic perturbations. In June 2025 it released X-Atlas/Orion, which it described as the largest publicly available genome-wide Perturb-seq atlas, covering about 8 million single cells across all human protein-coding genes with deep sequencing of more than 16,000 unique molecular identifiers per cell, roughly an eightfold jump over prior public datasets. The atlas was enabled by an internally developed, scalable screening platform the company calls Fix-Cryopreserve-ScRNAseq (FiCS) Perturb-seq, and was released under a non-commercial license alongside a bioRxiv preprint.
On March 17, 2026 Xaira launched X-Cell, described as its first virtual cell model and, at up to 4.9 billion parameters, the largest such model reported to date. X-Cell was trained on a successor dataset, X-Atlas/Pisces, comprising 25.6 million perturbed single-cell transcriptomes across seven biologically diverse cellular contexts (16 conditions in total), more than three times the scale of X-Atlas/Orion. Unlike the autoregressive single-cell models that had dominated the field, X-Cell uses a diffusion language model architecture that iteratively refines its predictions by progressively transforming control-state gene expression toward a perturbed state. Xaira reported that X-Cell was the first virtual cell model to demonstrate a clear scaling law, with perturbation-prediction accuracy following power-law improvements as data and model size grow, mirroring the scaling behavior seen in large language models. The company said the model generalized to previously unseen contexts such as iPSC-derived melanocyte progenitors and primary human T cells, and it made a subset of the Pisces dataset and the X-Cell model available to the research community. Xaira framed these models as tools for target and mechanism-of-action discovery, patient stratification, and toxicity prediction.
Xaira has emphasized building its AI and data platform first and allowing a therapeutic pipeline to follow from it, rather than licensing in a portfolio of preexisting drug candidates. Executives have described the company's experimental and clinical work as itself a vehicle for generating the data that improves its models, so that drug development and model development reinforce each other. Public statements point to therapeutic interest in oncology and in immunology and inflammation, with the Parker Institute for Cancer Immunotherapy among its backers, though the company has disclosed little detail about specific targets or development-stage programs. Tessier-Lavigne has said the ultimate measure of success will be whether the company's models can be used to make medicines, an acknowledgment that, as of 2026, Xaira's output to date has been platforms, datasets, and models rather than clinical candidates or approved drugs.
Operationally, Xaira maintained early teams in Seattle focused on foundational AI modeling and, in 2025, established and relocated its headquarters to the Gateway of Pacific campus in South San Francisco to expand its experimental capabilities while retaining the Seattle modeling group. The very large committed capital base is intended to let the company invest in expensive data generation and model training over a multiyear horizon without needing near-term revenue, a structure that distinguishes it from more conventional venture-funded biotech startups.
The financing was anchored by ARCH Venture Partners and Foresite Labs, with Foresite's Vik Bajaj and ARCH's Robert Nelsen closely involved in the company's creation and governance. The participating investor list spans crossover and venture firms active in both technology and life sciences, reflecting the company's positioning at the intersection of AI and biology.
| Fact | Detail |
|---|---|
| Public launch | April 23, 2024 |
| Committed capital | More than $1 billion |
| Lead investors | ARCH Venture Partners; Foresite Labs (Foresite Capital) |
| Other investors | F-Prime, NEA, Sequoia Capital, Lux Capital, Lightspeed Venture Partners, Menlo Ventures, Two Sigma Ventures, Parker Institute for Cancer Immunotherapy, Byers Capital, Rsquared, SV Angel |
| Incubated by | ARCH Venture Partners and Foresite Labs |
| Chief executive | Marc Tessier-Lavigne (ex-Genentech CSO; ex-Stanford and Rockefeller president) |
| Scientific co-founder | David Baker (UW Institute for Protein Design; 2024 Nobel laureate in Chemistry) |
| Other co-founders | Hetu Kamisetty (CTO, ex-Meta); Vik Bajaj (Foresite Labs CEO) |
| Chief medical officer | Paulo Fontoura (joined early 2025; ex-Roche) |
| Chief scientific officer | Debbie Law |
| Headquarters | South San Francisco, California (with Seattle modeling team) |
| Notable datasets / models | X-Atlas/Orion (about 8M-cell Perturb-seq atlas, June 2025); X-Cell virtual cell model with X-Atlas/Pisces (25.6M cells; up to 4.9B parameters, March 17, 2026) |
| Therapeutic focus | Oncology; immunology and inflammation (limited public disclosure) |
| Approved products | None as of 2026 |
Xaira's launch was widely cited as a marker of how far investor conviction in AI for biology had advanced by the mid-2020s, both for the unusual size of its initial raise and for the profile of its founders and board. By pairing the generative protein design lineage of the Baker lab with industrial-scale functional genomics data generation and large machine-learning models, the company became a prominent test of the thesis that combining better algorithms with proprietary, purpose-built biological data can make drug discovery faster and more predictive. Its public releases of large Perturb-seq datasets and the X-Cell virtual cell model also positioned it within the broader effort across the field to build "virtual cell" foundation models. As with other AI in healthcare ventures, the central open question remains whether platform and benchmark advances will translate into approved therapies; Xaira itself has stressed that this clinical validation is the standard by which it expects to be judged.