The new ‘virtual biopsy’ model predicts brain activity, therapy response, and side effects in patients with up to 3x greater accuracy than previous state-of-the-art models.

SAN FRANCISCO, June 16, 2026 /PRNewswire/ — Verge Labs, the frontier AI lab backed by Eli Lilly, BlackRock, and Y Combinator, today introduced vBx-1.0, a foundation model for precision neurology that dramatically outperforms the previous state of the art. Alongside a technical report, the company is now previewing the model with partner drug developers.

94 percent of neurology drugs fail in the clinic, often because the wrong drug is given to the wrong patient. In cancer, individual patients can be matched to the right drug because doctors can sample the tumor, analyze it directly, and pick the right treatment. In neurology, the brain cannot be sampled from a living patient.

From a routine blood draw, vBx-1.0 reconstructs the molecular state of a patient’s brain from blood with up to 3x greater accuracy than the previous state of the art.

The model also predicts how a patient’s disease will progress, and whether they will respond to a given therapy. Benchmarked on a large Parkinson’s Disease patient cohort, vBx-1.0 outperforms gold-standard clinical measures in predicting treatment response to levodopa, offering a 33% trial enrichment for responders, with the company estimating a 43% reduction in trial-size. Additionally, vBx-1.0 uncovered biologically interpretable signals in the brain that indicate why a given patient is likely to respond as they do.

Verge also benchmarked vBx-1.0 on data from its own Phase 1b trial of VRG50635, a PIKfyve inhibitor, in which approximately one third of ALS patients discontinued after their first dose, for unknown reasons. When vBx-1.0 ran zero-shot on baseline patient blood samples before treatment, it identified that the PIKfyve-driven biological program was already suppressed at baseline in the same patients who later could not tolerate the drug. The company estimates that using vBx-1.0 would allow them to screen out approximately 34% of such patients prior to enrollment.

vBx-1.0 is trained on Verge’s proprietary corpus of more than 12,000 brain transcriptomes across 6,500 patients, with matched single-cell, proteomic, genomic, and clinical data and a physical inventory of 900-plus frozen brain tissue samples.

“We increasingly think of brain tissue as ‘LiDAR for neuroscience,” said Alice Zhang, Verge’s Chief Executive. “Without it, every blood or imaging proxy is a flat picture, missing depth. Over ten years, Verge Labs has amassed a rich corpus of tissue data–enabling us to design a novel model architecture that links those proxies to what’s actually happening in the brain.”

vBx-1.0 is available to preview today through CONVERGE, and work is underway on collaborations to validate its performance on partners’ clinical trial datasets. Earlier this month the company announced its first platform partnership, a CNS collaboration with Tenacia Biotechnology. “If you have a trial that failed in a mixed population, or you’re designing one now, we want to run the virtual biopsy on it,” said Zhang.

Partners can reach the company at partnerships@vergelabs.com, and the technical report is available at https://vergelabs.com/resources/reports/vbx-one-technical-report.

About Verge Labs

Verge Labs is a frontier AI lab for human disease biology. Founded in 2015 and headquartered in San Francisco, the company builds world models of patient biology on proprietary multimodal human tissue data, and partners with drug developers to identify CNS targets, stratify patients, and predict treatment response. Learn more at vergelabs.com.

View original content to download multimedia:https://www.prnewswire.com/news-releases/verge-labs-unveils-vbx-a-breakthrough-foundation-model-for-precision-neurology-302800508.html

SOURCE Verge Labs