In the rapidly accelerating landscape of pharmaceutical research, a narrative has taken hold: Artificial Intelligence (AI) is the panacea for the drug discovery industry’s chronic inefficiency. From startups to Big Pharma, billions of dollars are flowing into machine learning models designed to sift through vast chemical libraries. Yet, in the Bay Area, Verseon Corporation is betting on a counter-intuitive premise. They argue that if you want to find something truly new, you must stop relying solely on the training data of the past.
Verseon has been operating since 2002, long before the current AI gold rush. While peers like Recursion and Exscientia have garnered headlines for their neural-network-driven pipelines, Verseon’s philosophy is rooted in the rigorous, first-principles application of molecular physics. Their mission is not just to accelerate discovery, but to navigate the vast, uncharted “chemical universe” that existing AI models are arguably programmed to overlook.
The Problem of Chemical Stagnation
To understand Verseon’s position, one must first look at the math of medicinal chemistry. For over 150 years, humanity has been synthesizing compounds—from the early days of chloral hydrate in 1869 to the massive, make-on-demand libraries of today, such as the ZINC-22 database, which boasts tens of billions of compounds.
However, Adityo Prakash, co-founder and CEO of Verseon, suggests this volume is deceptive. “If you cluster nearly identical structures into the same chemotype, the number collapses,” Prakash explains. He argues that the industry is trapped in a cycle of “relabeling”—making minor modifications to existing chemical backbones, such as swapping a fluorine atom for chlorine.
While current AI models are excellent at finding these incremental variations, they are inherently limited by their training data. If an AI is trained on what has already been done, it will logically suggest more of the same. Estimates suggest that the total number of theoretically synthesizable drug-like molecules is roughly 10³³. In this context, humanity’s cumulative output is not even a drop in the bucket; it is a few grains of sand. Verseon’s goal is to design the rest of the beach.
Chronology of a Physics-Based Approach
Verseon’s journey began in 2002, a full two decades before the explosion of generative AI. By focusing on computational platforms that simulate the fundamental physics of how atoms interact with protein pockets, the company sought to move beyond the limitations of traditional, library-based screening.
- 2002: Verseon is founded with a vision of using molecular physics to design small-molecule drugs from scratch.
- 2018: The company receives regulatory clearance in Australia to begin Phase I human trials for VE-1902, their flagship Precision Oral Anticoagulant (PROAC).
- 2019: Dosing for VE-1902 commences, marking a critical transition from computational design to human clinical safety validation.
- 2021: Verseon nominates VE-4840 as its primary candidate for diabetic eye disease, targeting plasma kallikrein to address underlying retinal vascular leakage.
- 2025/2026: Continued development of the pipeline, with new patent grants, such as the July 2026 European patent for their anticoagulant program, highlighting the ongoing evolution of their intellectual property.
Designing the Future: The Physics of Creation
Verseon describes its proprietary platform as "Deep Quantum Modeling." Unlike AI models that "predict" what a molecule should look like based on past success, Verseon’s system starts with the target protein.
“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?’” says Prakash. This is a design-centric approach, akin to how CAD and CAM revolutionized the aerospace and semiconductor industries. It is not about mining existing data; it is about building a bespoke solution for a biological lock.
The Anticoagulant Challenge
The primary test of this platform is the PROAC program. For decades, the pharmaceutical industry has struggled to develop anticoagulants that prevent dangerous clots without inducing significant bleeding risks. Warfarin, introduced in the mid-1950s, was notoriously difficult to manage, and its successors—the DOACs like Eliquis and Xarelto—still carry substantial bleeding profiles.
Verseon’s candidate, VE-1902, is a reversible covalent thrombin inhibitor. Its design goal is specific: block the thrombin activity that creates clots while leaving the platelet-activating role of thrombin intact. Preliminary data from rodent models and Phase I studies suggest that this separation of function could offer a near-normal bleeding profile, a “holy grail” in cardiovascular medicine. By enabling safer, long-term anticoagulation, Verseon hopes to allow for the combination of anticoagulants with antiplatelet therapies for patients who currently lack such options.
A Critical Lens on Mainstream AI
Verseon’s skepticism toward "AI-first" drug discovery is shared by a growing chorus of industry critics. Prakash distinguishes clearly between prediction and creation.
Systems like AlphaFold, while revolutionary for protein structure prediction, succeed because they interpolate within a massive, existing record of experimental data. However, the bottleneck in drug discovery has rarely been protein structure; it has been the creation of new, effective chemical matter.
Prakash notes that many AI-discovered candidates often mirror the structural patterns of existing drugs. He cites the early 2000s COX-2 inhibitors—Vioxx, Celebrex, and Bextra—as examples of the "same car with a new grille." He warns that many modern AI-designed compounds are simply "the same car with a new paint job."
This critique is supported by data. A 2022 assessment by the Chemical Abstracts Service (CAS) found that several early AI-designed clinical candidates lacked significant structural novelty, often clustering closely around the molecular shapes of established drugs. Similarly, a 2025 review found that ligand-based AI models were significantly more likely to produce low-novelty molecules compared to structure-based approaches.
Implications for the Future of Medicine
Verseon’s business model represents a departure from the "AI-in-a-box" strategy. Their platform uses AI, but only after the initial physics-based design and laboratory validation. Once a novel molecule is synthesized and tested, the company uses AI to iterate and generate variations, effectively closing the loop between the wet lab and the computer.
This hybrid approach acknowledges that while AI is an excellent tool for optimization—the “tweaking” of successful molecules—it remains a poor substitute for human-directed, physics-based innovation in unexplored chemical space.
The Pipeline and the Path Ahead
Today, Verseon boasts a robust pipeline of 14 named candidates across seven programs. These include:
- Cardiometabolic Diseases: Focused on stroke/heart attack prevention (VE-1902) and diabetic vision loss (VE-4840).
- Oncology: Targeting multidrug-resistant tumors, CD73-positive tumors, and metastasis, alongside novel chemotherapy agents.
While the pace of progress may seem measured—with some programs remaining in preclinical stages—Verseon’s supporters argue that the company is playing the long game. By avoiding the rush to push "AI-generated" clones through the clinic, they are building a portfolio of genuinely novel, proprietary compounds that have the potential to solve medical problems that have stymied the industry for over half a century.
As the hype cycle around generative AI in drug discovery matures, the industry will eventually face a reckoning: are these new drugs truly transformative, or merely iterative? If Verseon’s physics-first bets pay off, they may well prove that while AI can help us navigate the known world, only a deep understanding of physical law can help us discover the next one.
About the Author
