In the high-stakes world of pharmaceutical development, a quiet revolution has been unfolding—one that predates the current obsession with generative AI by nearly two decades. While Silicon Valley and Big Pharma scramble to integrate Large Language Models (LLMs) and neural networks into their discovery pipelines, the Bay Area-based company Verseon has maintained a singular, unwavering focus: the rigorous application of molecular physics to design drugs from the atomic level up.
Founded in 2002, long before "AI-driven drug discovery" became a buzzword, Verseon is challenging the industry’s reliance on data-driven prediction. Their central thesis is both simple and provocative: current AI models, while excellent at iterating on existing successes, are fundamentally incapable of true invention because they are constrained by the limited, biased training data of the past.
The Magnitude of the Challenge: A Counting Exercise
To understand Verseon’s unique position, one must first grasp the sheer scale of the chemical landscape. For over 150 years, since the introduction of chloral hydrate as a synthetic drug in 1869, medicinal chemists have synthesized approximately 10⁸ (100 million) drug-like compounds.
In the mid-2010s, databases like ZINC 15 attempted to catalog this "purchasable space," holding roughly 220 million molecules. By 2023, ZINC-22 pushed this further into the tens of billions. However, Verseon CEO Adityo Prakash argues that these numbers are misleading.
"Many of these are just relabelings of the same compound or minor structural tweaks—replacing a fluorine atom with a chlorine atom," Prakash explains. "If you cluster these nearly identical structures into their respective chemotypes, the number of truly distinct chemical entities collapses significantly."
The theoretical ceiling, by contrast, is astronomical. A 2013 study by researchers Pavel Polishchuk, Timur Madzhidov, and Alexandre Varnek estimated that the number of drug-like molecules that could theoretically be synthesized—given a cap of 36 heavy atoms—is roughly 10³³. To visualize this, if human history represents a few grains of sand, the total chemical universe is the entire planet.
"Think of the full set of possibilities as a chemical universe," Prakash says. "What humanity has explored so far is not even a planet. It is a few grains of sand."
Chronology: Two Decades of Computational Rigor
- 2002: Verseon is founded with a mission to develop computational platforms capable of de novo design, prioritizing physics-based simulations over statistical pattern matching.
- 2018: Verseon receives regulatory clearance in Australia to initiate a Phase I clinical trial for its lead precision oral anticoagulant, VE-1902.
- 2019: Dosing begins for VE-1902, targeting safe clot prevention without the traditional bleeding side effects.
- 2021: The company nominates VE-4840 as its primary candidate for diabetic retinopathy, moving into preclinical development.
- 2025/2026: Continued expansion of the pipeline, now covering seven distinct programs across cardiometabolic disease and oncology, with ongoing patent approvals, including a new European patent for its anticoagulant technology granted in July 2026.
Designing Beyond Known Chemical Space
Verseon’s approach is akin to the transition from manual drafting to Computer-Aided Design (CAD) and Computer-Aided Manufacturing (CAM) in aerospace and chip design. Instead of feeding a model a database of existing drugs and asking it to predict a new one—which leads to incrementalism—Verseon begins with the target protein.
"We start with a protein pocket 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?’" Prakash notes.
This physics-first approach allows them to venture into chemical regions that AI, which relies on the interpolation of existing datasets, would never reach. By ignoring the "training data" of past drug structures, Verseon aims to find scaffolds that are structurally distinct from anything currently in the pharmacopeia.
The Anticoagulant Frontier: Solving the Bleeding Risk
One of the most compelling applications of this technology is the search for an anticoagulant that does not increase bleeding risk. Since the 1950s, the pharmaceutical industry has struggled to improve upon warfarin, and while newer drugs like Eliquis (apixaban) and Xarelto (rivaroxaban) have improved dosing, they still carry significant bleeding risks.
Verseon’s Precision Oral Anticoagulant (PROAC) program aims to create reversible covalent thrombin inhibitors. The goal is to selectively block clot formation while leaving the platelet-activating role of thrombin intact.
The clinical stakes are high. Patients requiring both anticoagulants and antiplatelet therapy often face restricted treatment windows due to the additive risk of major bleeding. Data from the AFIRE study and recent 2025 meta-analyses suggest that monotherapy with safer anticoagulants could reduce major bleeding by up to 41%. Verseon believes that by sparing platelet activation, their VE-1902 candidate could redefine the standard of care for patients with atrial fibrillation.
Diabetic Eye Disease and Oncology
Beyond cardiology, Verseon is targeting high-unmet-need areas like diabetic macular edema. Current treatments, such as anti-VEGF injections (e.g., Eylea), are invasive and require frequent ocular administration. Verseon’s VE-4840, an oral plasma kallikrein inhibitor, aims to treat the underlying vascular leakage from within.
The company’s broader pipeline includes 14 candidates across seven programs. Their oncology research is particularly aggressive, focusing on multidrug-resistant tumors and CD73-positive cancer pathways, leveraging their physics-based engine to design novel chemotherapeutic agents that bypass traditional resistance mechanisms.
AI: The Utility of Interpolation vs. The Necessity of Extrapolation
Prakash is not anti-AI, but he is deeply critical of the "AI-first" dogma currently dominating drug discovery. He draws a firm line between prediction and creation.
"AI is good at interpolation and terrible at extrapolation," Prakash argues. "An AI system by itself will not hand you something fundamentally new."
He points to the industry’s history with COX-2 inhibitors like Vioxx and Celebrex, which shared almost identical chemical scaffolds. In the age of AI, this phenomenon is repeating. When models are trained on existing clinical candidates, they often produce "new" drugs that are essentially the same car with a different coat of paint.
A 2022 analysis by the CAS (Chemical Abstracts Service) regarding Exscientia’s clinical candidates found that many claimed molecules shared the exact same structural "shapes" as previously approved drugs, leading the organization to suggest that the structural innovativeness of these AI-designed candidates "might not set the world on fire."
Verseon’s methodology flips the script. AI is utilized only after the physics-based engine has proposed a novel, synthesized structure. The AI then assists in the optimization process—creating variants to refine potency and safety—rather than defining the starting point.
Implications for the Future of Medicine
As the industry matures, the distinction between "physics-based" and "data-based" drug design will likely define the next generation of biotech success. While companies like Insilico Medicine are achieving milestones—such as their Phase 3 trial for the TNIK inhibitor rentosertib—the underlying question remains: how much of the future of medicine should be based on the patterns of the past?
Verseon’s persistence suggests that while AI can accelerate the refinement of drugs, the innovation of drugs requires a deeper understanding of the physical reality of molecular interaction. If their candidates, such as VE-1902 or VE-4840, successfully reach the market, they will provide a powerful validation for the company’s "Deep Quantum Modeling" approach.
For the pharmaceutical industry, the implications are clear: the "low-hanging fruit" of chemical space has been picked. To treat the most complex diseases, we can no longer rely solely on the digital ghosts of past research. We must return to the fundamental laws of nature, building new, bespoke molecules from the ground up—one atom at a time.
About the Author
