Bexorg
Bexorg tests drugs for central nervous system diseases in donated postmortem human brains before they reach clinical trials, working for pharmaceutical and biotechnology sponsors and on its own licensed candidates. Its BrainEx platform, built on technology the founders developed at Yale, perfuses whole human brains with a custom artificial blood that restores metabolic and molecular activity for prolonged periods. A candidate is delivered through the brain's own vasculature and blood brain barrier, whether it is a small molecule, a biologic, a viral vector or a lipid nanoparticle, so a team can measure exposure, target engagement, pharmacokinetic and pharmacodynamic relationships and biomarkers in human tissue. NeuroScale extends the work to brain slices, and NeuroLens is the AI layer, a foundation model trained on the transcriptomic, proteomic and metabolic data the experiments generate, used to model drug response and point to targets and biomarkers.
Biohaven began a two program collaboration in 2025. BrainEx showed its PKM2 modulator BHV-8100 crossing the blood brain barrier and tripling glucose utilization in human brains, data that supported the investigational new drug filing, and Biohaven dosed the first human participants in June 2026. The Michael J. Fox Foundation funds a separate Parkinson's biomarker project using tissue from LifeNet Health, and in October 2026 ARPA-H awarded Bexorg up to $28.3 million to extend how long brains can be studied.
Zvonimir Vrselja, the chief executive, and Nenad Sestan founded Bexorg in 2021 as a Yale spinout. It runs a laboratory in New Haven with an office in San Francisco. An October 2025 Series A of $23 million, led by Engine Ventures with Amplify Partners, Starbloom Capital, Connecticut Innovations and E1 Ventures, brought total funding to $42.5 million.
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
BrainEx supplies the experimental measurements for each program, including drug exposure and target engagement in donated human tissue, and NeuroScale adds a slice based format. NeuroLens is the computational part, learning from molecular and experimental data to produce hypotheses about targets, biomarkers and drug response. The experiments produce measurements a sponsor can use without the model, as the Biohaven results show.
The model adds predictions to test with those experiments and to prioritize later work. The October 2025 Series A was raised to expand that AI engine on whole human brain data, so the AI layer is growing, while the perfusion platform remains the core of the offer.
Experimental scientists and drug sponsors use the platform to decide which findings deserve more work. A sponsor can commission specific measurements, such as whether its candidate reaches a target in a relevant tissue type, and NeuroLens prioritizes findings a program could advance or test again.
People decide at each step, but no handoff from model recommendation to sponsor decision is defined, and there is no numeric uncertainty threshold or required second review for a candidate screen. Biohaven keeps responsibility for its filings and clinical development, and the ethics board oversees research involving donated tissue rather than model output, so oversight of the AI layer rests on the scientists using it.
NeuroLens combines molecular measurements and observations gathered over the course of human brain experiments, covering gene expression, proteins and metabolism, and the Biohaven collaboration describes transcriptomic, proteomic and metabolic readouts in disease relevant states. Bexorg trains a foundation model on that human brain molecular dataset, and the model supports predictions about drug response and potential biomarkers.
The source measurements and the purpose of the learning system are identified. The model architecture, released version, training and evaluation split and change schedule are not given, and a sponsor receives results tied to its commissioned program rather than access to the model itself.
Discovery runs on proprietary experimental data and the NeuroLens computational work, and the company describes a foundation model trained on data its own research platform generates. University, foundation and pharmaceutical relationships extend the experiments and the drug programs. No supplier is named for any underlying model or for the infrastructure used to train or serve the computational system, and the architecture is called proprietary without further detail.
How internal models relate to any third party components, and how a partner would hear about a change to an underlying model during a program, are not stated, so the chain behind a prediction cannot be traced past the company.
The 2025 Biohaven collaboration covered two preclinical programs, and BrainEx produced the clearest result: BHV-8100 crossed the blood brain barrier and tripled glucose utilization in donated human brains, with sustained effects in brains from donors with Alzheimer's disease and dementia. Those data supported the investigational new drug filing, and Biohaven began first in human dosing in June 2026.
The founders' 2019 Nature study restored cellular and metabolic activity in postmortem pig brains and grounds the experimental method. None of these results measures how accurately NeuroLens predicts clinical outcomes, which the ARPA-H project begins to test by comparing BrainEx observations with known clinical outcomes.
The experimental program uses tissue donated for research with authorization from donors or their families. The New Haven laboratory may keep some tissue for future research and disposes of material it will not use. NeuroLens learns from data generated from donated tissue, so model training is a defined use of the experimental data, and the 2025 Series A was raised to expand that AI engine, so the training set grows with each program.
The tissue process is described in more detail than the data process. Deidentification of linked donor records, default retention for derived records and how material or data from a pharmaceutical partner is kept separate are not explained.
The platform serves research collaborations built on a sponsor's compounds and study questions, generating molecular data from donated tissue, sometimes in a program with a named procurement organization or university partner. The research FAQ covers donor authorization. Whether linked health records or identifiers enter partner projects, which parties handle protected information, and whether Bexorg acts as a business associate or signs a business associate agreement are not stated.
Donor clinical histories, such as an Alzheimer's or dementia diagnosis, are part of what makes the tissue useful, so the handling of that information matters even in preclinical work. Tissue consent sets the scope of the donation, and the rest falls to each partner's own agreement.
Bexorg runs the experimental platform and manages the computational analysis for each research program, and its independent bioethics committee oversees research practice, including how tissue is authorized and handled. No information security audit, certification or trust center covers the data and model systems, and none of the named partnerships comes with a scoped attestation for the systems a commercial partner's data would touch.
The lab and leadership are identified, but how access to study data is controlled and audited for each sponsor is not described. For pharmaceutical partners sharing unreleased compounds, that leaves confidentiality resting on contract terms alone.
BrainEx and NeuroLens support preclinical discovery and translational research, generating experimental evidence and hypotheses for drug programs through research collaborations. Biohaven used BrainEx data to support its investigational new drug filing for BHV-8100, and that filing belongs to Biohaven's candidate and its move into a first in human study in June 2026.
The platform makes no diagnosis or treatment recommendation for patients, and no device clearance is claimed, which fits a research service. The ARPA-H project will compare platform observations with known clinical outcomes as further research, and the award does not give the platform a regulatory status.
An independent bioethics board oversees research practices, with members who bring expertise in donation, law and research integrity, and tissue comes from authorized donors with a range of disease histories, including Alzheimer's and Parkinson's disease. The Parkinson's biomarker project sources donated tissue through LifeNet Health. Those are specific sourcing and oversight practices for the experimental side.
The computational system has no model card, no process for assessing model changes and no performance results by cohort. The bioethics board's remit covers research practice, not how NeuroLens behaves across donor groups that differ by age, sex, disease stage or time after death.
A prediction about a target or biomarker can be tested against measurements from the experimental platform, which gives a sponsor a way to check the model's output inside the same program. In the Biohaven collaboration, the sponsor used the research results to decide how to advance its own candidate, and Biohaven, not Bexorg, carries responsibility for the filings and clinical development that follow. Partnership questions go through a general company contact.
There is no formal process for correction, investigation or response times when a model result is disputed, and no outcome warranty or benchmark for prediction errors. How study findings are checked and how a discrepancy is resolved depends on each research agreement.
No clinical record integration exists, which fits a preclinical service, so the question is how results reach a sponsor's own research systems. Partners receive measurements of target engagement, compound exposure and pharmacology from programs that originate in Bexorg's experimental systems. Biohaven is the named preclinical partner, and LifeNet Health supplies tissue for the Parkinson's project.
No sponsor laboratory system, research data platform or standard interface for moving measurements into a research environment is named. The collaboration exchanges scientific results without stated connectors, transport protocols or a partner data pipeline, so each program's data handoff is set case by case.
Tissue research runs at the company's laboratory in New Haven, Connecticut, where experimental measurements and computational analysis come together inside the program run for a sponsor. Partners buy the discovery work under a research agreement rather than installing software in their own facilities, so deployment is a service model. The physical lab location is known, but the location and operator of the systems used for model training and electronic data storage are not stated. There is no choice of cloud region and no partner managed deployment for the computational work, which matters for a sponsor that wants derived data held in its own environment.
The commercial offer is a research collaboration built around a sponsor's discovery questions. Biohaven's agreement covers two candidates and sets scientific objectives that include target engagement, biomarkers and pharmacokinetic and pharmacodynamic relationships, and new work starts through a company contact. There is no standard rate per experiment, candidate or program, and the Biohaven deal economics are not disclosed.
The public amounts, a $23 million Series A and an ARPA-H award of up to $28.3 million, finance the company and a specific research project rather than set a price. A drug developer has no rate it can apply to its own scope before that first conversation.
Central nervous system drug development is the focus, and the platform uses tissue from donors with Alzheimer's or Parkinson's disease and from donors with documented dementia to study exposure and target engagement before human trials. Small molecules, biologics, viral vectors and lipid nanoparticles can all be delivered through the brain's vasculature.
Biohaven's BHV-8100 program, aimed at metabolic and immune dysfunction in neurodegeneration, is the clearest drug development example, the Michael J. Fox Foundation backs a Parkinson's biomarker project, and an Oxford and Medical Research Council collaboration works on translational gene therapy. Those programs span several research settings, while validation of the AI models across diseases and drug modalities is still limited.
Alternatives to Bexorg
Other products that do the same kind of work as Bexorg, and what each one does.
Clinical stage biotechnology company now trading as Verge Labs, applying machine learning to human tissue derived multiomic data in neurology.
Molcure designs antibodies and peptides for drug companies, using a large language model for antibodies that it built from scratch together with physics based simulations and experimental data from its own biolab.
Insilico Medicine is a generative AI drug discovery company (HKEX: 3696), headquartered in Cambridge, Massachusetts with operations in Hong Kong, Abu Dhabi and Shanghai, that both licenses software and develops its own medicines.
Numerion Labs, formerly known as Atomwise, is a San Francisco AI drug discovery company developing small molecules for immune and inflammatory disease, led by chief executive Steve Worland.
Variational AI develops Enki, a generative model for designing small molecules against a research team's target product profile.
Chemify makes the molecules it designs, in a laboratory it built itself. Rather than handing a design to somebody with a laboratory, it is trying to turn its own laboratory into a printer.
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 |
|---|---|---|---|---|
|
Contact for pricing
|
Research collaboration negotiated by program | No business associate agreement is stated for research collaborations | No setup or program fee is listed. | Vendor Published |
Bexorg works with drug developers through individually negotiated research agreements. The Biohaven collaboration covers two preclinical programs, and no standard fee or deal terms are given.