Biotechnology · AI drug discovery
Verge Genomics
The first company to carry a drug from an AI platform into ALS patients. The biomarker moved the wrong way, the program was dropped, and in May 2026 the company stopped developing drugs to sell the data instead.
The number that went the wrong way
Fifty four people with ALS took a capsule every day and gave blood. Verge was watching one number: neurofilament light chain, a protein that spills out of neurons as they die. Lower is better. The study was powered to detect a 30 percent drop.
Within two weeks the number went up. It stayed up for as long as patients stayed on the drug.
The company later worked out why. VRG50635 blocks an enzyme called PIKfyve, which sits in the machinery cells use to clear their own waste. Inhibiting it opens a route that pushes intracellular contents out. The drug was not failing to reach the brain. It was reaching the brain and flushing the contents of sick cells into the bloodstream, neurofilament among them.
“Plasma NfL was tracking the drug’s action, not the disease. Unfortunately, it was also our efficacy endpoint,” Alice Zhang wrote in May 2026.
Why ALS breaks drug companies
ALS kills motor neurons and, typically, the patient within two to five years of diagnosis. About nine in ten cases are sporadic, with no inherited cause to aim at, and no animal model reliably predicts what a compound will do in a person. So the field leans on biomarkers, and neurofilament became the one everybody used. Verge built its trial around it.
The trial taught the field something it did not want to learn. When a drug engages the cell’s recycling system, blood neurofilament can move for reasons unrelated to whether neurons are being saved. Spinal fluid told a different story in the same patients: CSF neurofilament trended down, more so in the C9orf72 genetic form. Plasma went the other way.
The founder who left medical school three months short
Alice Zhang graduated from Princeton, won a Paul and Daisy Soros Fellowship in 2012, and entered the MD and PhD program at UCLA. She left three months before graduating and started Verge in 2015 with Jason Chen, whom she met in the program.
“I just became very frustrated with the drug discovery process,” she said at the time. “It’s largely a guessing game where companies are essentially brute force screening millions of drugs just to stumble across a single new drug that works.”
The company went through Y Combinator in the summer of 2015 and raised a $32 million Series A in July 2018, led by DFJ with WuXi AppTec’s venture fund, ALS Investment Fund, Agent Capital and OS Fund.
Not everyone was persuaded. That same month the chemist Derek Lowe used his Science blog to file an objection that has aged into something close to a prophecy. “The problems with Alzheimer’s, ALS, and Parkinson’s drug discovery are not data handling problems,” he wrote. “But what we’re short of is insights, ideas, and understanding, and those come slowly, painfully, and expensively.”
The money and the partners came first
In July 2021 Eli Lilly paid $25 million upfront for an ALS discovery collaboration with up to $694 million in milestones behind it. In December 2021 BlackRock led a $98 million Series B, joined by Lilly, Merck Global Health Innovation Fund, Section 32 and Vulcan Capital. In September 2023 AstraZeneca’s Alexion signed a four year deal worth up to $42 million in upfront, equity and near term payments plus up to $840 million in milestones. In March 2024 Ferrer took ex US rights to VRG50635 for more than 112.5 million euros.
Every one of those deals was signed before a single ALS patient result existed.
A no go decision, then a new company
Enrollment finished in August 2024. The registry records the study as completed on July 7, 2025. On December 5, 2025 the record was changed to terminated, with the reason given as “lack of risk-benefit data.”
On December 18, 2025 Fierce Biotech reported that Verge was dropping its only clinical asset. Its vice president of operations, Effie Bruzik, said the drug had failed a pre-specified efficacy analysis and “A no-go decision was made.” She pointed at a different market: “Recent breakthroughs in AI foundation models has meaningfully increased demand for Verge’s multi-modal human datasets in the last 18 months, and we see a clear opportunity to drive industry-wide impact by enabling partners to improve pipeline success rates.”
On May 27, 2026 the company relaunched as Verge Labs. Roughly 90 percent of staff were let go. Four senior hires arrived from Altos Labs, Calico, PostEra and Flatiron Health. In June 2026 it named a first platform customer, Tenacia Biotechnology, and released vBx-1.0, a model that claims to reconstruct brain biology from a blood draw. The legal entity, Verge Analytics, Inc., never changed.
What is proven, and what is still claimed
| Evidence | What the record shows | Source type |
|---|---|---|
| Phase 1b in ALS | NCT06215755. Open label, 54 patients, Jan 15, 2024 to Jul 7, 2025. Status TERMINATED, reason “lack of risk-benefit data.” No results posted as of Sep 22, 2026. | Public record |
| Which endpoint failed | All six registered primary outcomes are safety measures, from adverse events to ECG abnormalities. Plasma neurofilament change is registered as a secondary outcome. The company says the drug “missed its primary efficacy endpoint.” | Differs from registry |
| When it ended | Registry actual completion: Jul 7, 2025. The company says late 2025; BioPharma Dive reported it wrapped in December. | Differs from registry |
| The 2023 Phase 1 | A June 2023 release describes a Phase 1 in 80 healthy volunteers. No registration for it was found. A sponsor search on Sep 22, 2026 returns two studies only, the other a 22 person study run Feb to Mar 2024. | Not found |
| Target engagement | Peer reviewed. GPNMB validated as a pharmacodynamic biomarker of PIKfyve inhibition, Clinical and Translational Science, Feb 2026. The drug did reach the brain and hit the pathway. | Public record |
| Digital endpoints | Peer reviewed. Speech based digital endpoints tracked ALS progression in the VRG50635 trial, Scientific Reports, Apr 14, 2026. | Public record |
| Platform output | More than 280 novel targets, an 83 percent preclinical validation rate, $1.6 billion in partnerships, two AI discovered clinical candidates. None independently audited. One reached patients and did not show benefit. | Company-stated |
| Dataset size | More than 12,000 brain transcriptomes across 6,000 patients in the May 2026 release, and across 6,500 in the June 2026 releases. Not verified independently. | Company-stated |
| Ferrer partnership | Fierce Biotech reported the March 2024 license was left in limbo once the program was dropped, and that neither company answered questions. No update since. | Unresolved |
| Federal record | No FDA device records, as expected for a drug developer. No NIH or NSF awards to the company found. | Not found |
Read plainly: the platform did something real and something incomplete at once. It nominated a target nobody had prosecuted in ALS, the chemistry reached the brain, and the pharmacodynamic biomarker moved as predicted. That is more biological signal than most ALS trials produce. What it could not do was say which patients to enroll, or which way the efficacy biomarker would move once the drug was on board. Closing those two gaps is what the new company says it is for, and neither has been closed in public.
What to watch
- Whether results from NCT06215755 are ever posted. Nothing has appeared in fourteen months.
- The status of the Ferrer license, and whether any of its announced value survives.
- Independent validation of vBx-1.0. Its accuracy claims rest on the company’s own benchmarks.
- A financing for the new model. The last Form D on record was filed in January 2022.
In their words
“I would love to be the company that puts the first drug from a computational platform into a human and shows that it actually slows or reverses disease.”
Alice Zhang, co-founder and CEO, Threshold Ventures founder Q and A, 2018 · Interview
“But what we’re short of is insights, ideas, and understanding, and those come slowly, painfully, and expensively.”
Derek Lowe, In the Pipeline, Science, July 2018 · Independent
“The lack of predictive animal models is one of the greatest challenges to developing effective ALS therapies today.”
Alice Zhang, Phase 1 results release, June 2023 · Company release
“We are thrilled to be the first company to evaluate multiple digital clinical biomarker endpoints in an ALS therapeutic trial to gauge short-term disease progression across key functional domains.”
Diego Cadavid, chief medical officer, April 2025 release · Company release
“Never in history has a transformational technology been just built overnight, and it is very rarely the case that the first attempt is the blockbuster.”
Alice Zhang to BioPharma Dive, May 2026 · Independent
“For a decade, Verge has been building the human tissue data that our field has been missing.”
Alice Zhang, Verge Labs launch release, May 2026 · Company release
“We got very close. We were in the right neighborhood. We just have not yet found the right house yet.”
Alice Zhang, Verge Labs blog, May 2026 · Company
Related companies
Sources
- Public recordClinicalTrials.gov NCT06215755, Phase 1b of VRG50635 in ALS
- Public recordGPNMB, a pharmacodynamic biomarker for PIKfyve inhibition with VRG50635
- Public recordForm D, Verge Analytics, Inc., CIK 0001894889
- Public recordHercules Capital schedule of investments, Verge Analytics lines
- IndependentThe Case of Verge Genomics
- IndependentVerge signs on Merck, Lilly for nearly $100M round
- IndependentAstraZeneca taps a genomics biotech and its AI technology
- IndependentFerrer signs $122M biobucks deal for ex-US license
- IndependentVerge Genomics drops sole clinical candidate
- IndependentVerge, following trial failure, rebrands its ambitions
- InterviewAlice Zhang, Verge Genomics, founder Q and A
- CompanyVerge Genomics Secures $98 Million in New Financing
- CompanyPhase 1 safety and tolerability data for VRG50635
- CompanyWhat We Learned From The Trial of Our First AI-Discovered Drug
- CompanyVerge Labs Unveils vBx, a model for precision neurology
Profile researched and written by Healthcare Discovery. Last updated September 29, 2026.
