Bay Area-based pharmaceutical company Verseon has been a quiet pioneer in computational drug design since 2002, predating the widespread emergence of AI drug discovery firms like Recursion and Exscientia and two decades before the public launch of generative AI tools such as ChatGPT. Today, Verseon asserts that its proprietary computational platform yields novel drug candidates fundamentally unobtainable through any other existing methodology, setting it apart in an increasingly AI-dominated landscape.
The core of Verseon’s ambitious claim rests on a profound understanding of chemical space—the theoretical universe of all possible molecules. While chemists have diligently explored drug-like compounds for over 150 years, starting with the clinical introduction of chloral hydrate in 1869 as the first synthetic drug, the vastness of unexplored chemical possibilities remains staggering. Contemporary enumerated screening collections, such as ZINC 15 (2015) with its 220 million molecules, have grown exponentially, with ZINC-22 (2023) now referencing make-on-demand libraries containing tens of billions of compounds.
However, as Adityo Prakash, Verseon’s co-founder and CEO, points out, this numerical growth can be deceptive. “Many are relabelings of almost the same compound or tiny changes around the same chemical backbone, such as replacing fluorine with chlorine,” Prakash explains. “If you cluster nearly identical structures into the same chemotype, the number collapses.” This highlights a critical limitation: the vast majority of explored chemical entities are mere variations of known structures, tethering drug discovery to an already familiar, albeit large, territory.
Navigating the Infinite Chemical Universe
The true scale of the chemical universe dwarfs even the largest databases. A seminal 2013 estimate by Pavel Polishchuk, Timur Madzhidov, and Alexandre Varnek, extrapolating from constrained molecular graph enumeration and capping drug-likeness at 36 heavy atoms, suggested that the number of synthesizable drug-like molecules could be as high as 10^33. Earlier, broader estimates ranged from 10^23 to 10^60. To put this into perspective, the estimated number of atoms in the observable universe is around 10^80.
“Think of the full set of possibilities as a chemical universe,” Prakash urges. “What humanity has explored is not a planet. It is a few grains of sand.” This immense, largely unexplored chemical space represents both the greatest challenge and the greatest opportunity in drug discovery. Traditional high-throughput screening and even most AI-driven approaches are inherently limited by their reliance on existing data, meaning they predominantly search within or interpolate from these "grains of sand." Brute-force exploration of such a massive space is simply impossible.
Verseon’s De Novo Design Philosophy: Physics at the Atomic Level
Against this backdrop, Verseon champions a fundamentally different approach: de novo drug design rooted in molecular and quantum physics. “We need computational methods that design new things, the way CAD and CAM transformed buildings, airplanes and advanced computer chips,” Prakash states. Verseon’s platform, dubbed "Deep Quantum Modeling," begins not with existing molecules, but with a target protein pocket. The system then employs sophisticated physics-based calculations to construct an entirely novel chemical structure, atom by atom, ensuring it fits precisely into the target and forms the desired chemical interactions.
“We start with a protein and ask, ‘Can I create a completely novel chemical structure that humanity has never made, fit it into this pocket and arrange the atoms so it binds and forms the right chemical interactions?’ That is what we do,” Prakash elaborates. This process is distinct from conventional virtual screening, which sifts through libraries of known or readily synthesizable compounds, and also from many generative AI models that typically learn patterns from existing molecular data to propose new ones that are often structurally similar. Verseon’s methodology prioritizes true novelty and precise atomic-level engineering, aiming to unlock entirely new chemotypes.
A Pipeline Targeting Unmet Medical Needs
Verseon’s pipeline reflects this commitment to novel chemistry, spanning critical therapeutic areas such as cardiometabolic disorders and cancer. The company lists seven programs and fourteen named candidates, typically with two candidates per program to provide multiple strategic shots on goal within each disease area. These include candidates for stroke and heart attack prevention, diabetic vision loss, hereditary angioedema (HAE), fatty liver disease, multidrug-resistant cancers, CD73-positive cancers, and metastasis. “Our pipeline is filled with candidates that have properties current methods have struggled to produce,” Prakash affirms.
The Holy Grail: An Anticoagulant Without the Bleeding Risk
One of Verseon’s most compelling programs is its Precision Oral Anticoagulant (PROAC) initiative, aimed at preventing strokes and heart attacks. The challenge of developing anticoagulants that effectively prevent dangerous clots without causing excessive bleeding has plagued the pharmaceutical industry for decades. Warfarin, discovered in 1939 and commercially launched as a rodenticide in 1952 before human use in the mid-1950s, was a significant breakthrough but came with a narrow therapeutic window and numerous drug-food interactions.
The current generation of direct oral anticoagulants (DOACs) like Eliquis (apixaban), Pradaxa (dabigatran), and Xarelto (rivaroxaban) have largely replaced warfarin due to their improved safety profile and convenience. However, they still carry a significant bleeding risk. “Drugs such as Eliquis, Pradaxa and Xarelto still carry too much bleeding risk,” Prakash states. This risk is particularly problematic for patients requiring concomitant antiplatelet therapy, where guidelines often limit the duration of combined treatment due to increased major bleeding events. For instance, the AFIRE trial showed major bleeding at 1.62% per patient-year on rivaroxaban alone, rising to 2.76% with an antiplatelet added. A 2025 meta-analysis highlighted a 41% reduction in major bleeding on anticoagulant monotherapy.
Verseon’s PROAC compounds are reversible covalent thrombin inhibitors, meticulously designed to selectively block clot formation while preserving thrombin’s crucial platelet-activating role. The company posits that this precise separation of functions is key to maintaining a near-normal bleeding profile. On July 15, the European Patent Office granted Verseon a new patent covering this innovative program. Verseon believes that such a thrombin inhibitor, sparing platelet activation, could eventually be safely paired with an antiplatelet drug for much longer durations than currently feasible, offering a significant therapeutic advantage.
The lead candidate, VE-1902, has progressed to human testing. In September 2018, Verseon received regulatory clearance in Australia for a double-blind, randomized, placebo-controlled Phase I study in healthy volunteers, focusing on safety, tolerability, and a composite hemostatic profile, with pharmacokinetics and pharmacodynamics as secondary endpoints. Dosing commenced in early 2019, and the compound remains listed at Phase I on Verseon’s pipeline page. Preclinical data published in a 2020 paper in Thrombosis Research demonstrated that VE-1902 produced antithrombotic effects with significantly less bleeding compared to comparator anticoagulants in rodent models, reinforcing its potential.
An Oral Solution for Diabetic Eye Disease and Other Endeavors
Beyond anticoagulation, Verseon is also making strides in diabetic eye disease, an area with substantial unmet needs. Patients suffering from vision-threatening diabetic macular edema (DME) often require repeated intravitreal injections of anti-VEGF agents like Avastin (bevacizumab) or Eylea (aflibercept). These treatments, while effective, place a significant burden on patients due to the invasive nature and frequency of administration.
Prakash highlights Verseon’s unique approach: “We have developed oral drugs aimed at the underlying fluid leakage that damages the back of the eye.” Verseon’s compounds target and inhibit plasma kallikrein, a protein well-established as a key mediator in retinal vascular permeability. An oral therapeutic that addresses the root cause of fluid leakage could revolutionize the management of diabetic retinopathy and DME, offering a less invasive and more convenient treatment option.
In August 2021, Verseon nominated VE-4840 as its primary diabetic retinopathy candidate, citing promising results in a rodent model that showed reduced diabetes-induced retinal vascular permeability, alongside favorable preliminary toxicology data. While the compound is currently listed as preclinical, its potential to shift treatment paradigms from chronic injections to an oral regimen is considerable.
Other programs in Verseon’s diverse pipeline include treatments for hereditary angioedema (HAE), fatty liver disease (NASH/NAFLD), and several distinct cancer indications. In cancer, the company is developing novel chemotherapy agents, as well as therapies for multidrug-resistant tumors, CD73-positive tumors (an immune checkpoint target), and metastasis. Each of these programs aims to introduce fundamentally new chemistries to address challenging disease mechanisms.
AI’s Role: A Tool for Interpolation, Not Always for True Creation
While Verseon leverages advanced computational techniques, its leadership offers a nuanced and often critical perspective on the mainstream AI drug discovery field. Prakash draws a crucial distinction between prediction and creation, particularly when it comes to novel chemical matter.
He acknowledges AI’s successes in areas like protein folding, citing systems such as AlphaFold. These systems excel at predicting protein structures because they interpolate within a dense, experimentally rich dataset. “Researchers will still want external experimental data to validate some predictions, or closely related data that makes a prediction more reliable,” Prakash notes. He also points out that protein structure prediction, while valuable, “was never the major bottleneck in drug discovery,” with decades of experimental structural work forming the foundation for most programs. AI, in this context, fills gaps and offers predictions that researchers cautiously validate.
However, Prakash contends that the harder problem—creating genuinely new chemical entities in regions of chemical space with little or no relevant training data—is where AI’s limitations become apparent. “AI is good at interpolation and terrible at extrapolation,” he argues. “An AI system by itself will not hand you something fundamentally new.”
When trained on existing compounds and tasked with generating new ones, AI models tend to produce variations on what is already known. Prakash, echoing his head of chemistry Kevin Short, uses a compelling analogy: in the early 2000s, drugs like Bextra, Celebrex, and Vioxx were selective COX-2 inhibitors sharing a common mechanism and similar diaryl heterocycle scaffolds. These, he suggests, were like "the same car with a new grille." Many AI-discovered compounds, in Short’s analogy, are merely "the same car with a new paint job."
This isn’t to say AI is without value. Prakash concedes, “Could AI suggest that an old drug might work for another indication? Absolutely. AI is useful in those settings.” AI excels at finding patterns, optimizing known scaffolds, and identifying potential repurposing opportunities within existing data.
Independent analyses support some of these observations. A 2022 study by CAS (Chemical Abstracts Service) examined the structural innovativeness of Exscientia’s first three clinical candidates. It found that a significant proportion of these molecules shared molecular shapes with previously known or approved drugs (e.g., DSP-1181 with haloperidol, DSP-0038 with approved antipsychotics). CAS concluded that the candidates’ structural innovativeness “might not set the world on fire,” advocating for AI-designed molecules to meet the same novelty standards as those designed by medicinal chemists.
Further supporting this, a 2025 meta-analysis of 71 published cases revealed that molecules generated by ligand-based AI models often exhibited relatively low novelty, with 58.1% having a maximum Tanimoto similarity above 0.4 to existing compounds. In contrast, structure-based approaches showed lower similarity (17.9%). The authors cautioned that fingerprint metrics, while useful, can sometimes miss scaffold-level similarities, meaning a molecule might appear dissimilar while retaining a familiar core structure.
The Hybrid Future: Physics-First, AI-Optimized
Despite these critiques, the AI drug discovery field continues to advance, demonstrating its potential. Insilico Medicine, for example, has announced a Phase 3 trial for rentosertib, an oral TNIK inhibitor for idiopathic pulmonary fibrosis. Its target was identified using AI, and its molecular structure was generated via the company’s generative chemistry platform. Insilico went public on the Hong Kong Stock Exchange in December 2025. While TNIK and earlier inhibitors were known, and antifibrotic effects of TNIK inhibition reported in animal models, rentosertib reportedly uses a distinct core scaffold and binding mode compared to earlier inhibitors, showcasing AI’s ability to find new paths to known targets.
Verseon’s integrated approach marries its foundational physics-based design with targeted AI application. The company’s Deep Quantum Modeling platform initiates the de novo design of entirely new chemical structures. After these proposed molecules are synthesized and experimentally tested, AI is then introduced. At this stage, AI proposes variations based on the new experimental data generated by Verseon’s novel compounds. Scientists then critically evaluate which of these AI-suggested compounds to synthesize next.
“The AI learns from the new biological data and does what it is supposed to do: create variants,” Prakash explains. “Scientists still have to make the proposed molecule in the lab and validate the prediction. The AI helps with the tweaking process.” This strategy positions AI as a powerful optimization and exploration tool after fundamental novelty has been established through physics-driven design and initial experimental validation.
Verseon’s distinctive methodology represents a significant divergence from the predominant AI-first narrative in drug discovery. By prioritizing fundamental physics and de novo design to explore truly uncharted chemical territory, and then strategically employing AI for optimization, the company aims to deliver therapeutics that are not merely improvements on existing drugs but are genuinely novel and uniquely suited to address long-standing medical challenges. This approach underscores a future where diverse computational strategies, each playing to its strengths, collectively accelerate the discovery of life-changing medicines.
