In the fast-paced world of biotechnology, where artificial intelligence (AI) has become the industry’s favorite buzzword, a Bay Area-based firm is taking a contrarian stance. Verseon, which has been quietly refining its computational platform since 2002—long before the current generative AI gold rush—is betting that the future of drug discovery lies not in training models on past data, but in the fundamental principles of molecular physics.
As the industry grapples with the limitations of "AI-first" approaches, Verseon’s CEO, Adityo Prakash, argues that while machine learning is an excellent tool for optimization, it is fundamentally ill-equipped to invent the breakthrough medicines of tomorrow.
The Limits of the Chemical Universe
To understand Verseon’s mission, one must first grasp the sheer scale of the "chemical space" that researchers attempt to navigate. For over 150 years, since the clinical introduction of chloral hydrate in 1869, medicinal chemists have been building a library of drug-like compounds. By the mid-2010s, collections like ZINC 15 cataloged roughly 220 million molecules. Today, with the advent of "make-on-demand" libraries, that number has ballooned into the tens of billions.
However, Prakash cautions that quantity does not equal innovation. "Many of these are just relabelings of the same compound or tiny modifications around a familiar chemical backbone—like replacing a fluorine atom with chlorine," he explains. When these nearly identical structures are clustered into "chemotypes," the perceived vastness of modern drug discovery collapses into a narrow corridor of well-trodden paths.
Scientific estimates suggest that the number of theoretically synthesizable, drug-like molecules could reach as high as 10³³—a staggering figure that makes current human exploration look like a few grains of sand on a vast beach. "Think of the full set of possibilities as a chemical universe," Prakash says. "What humanity has explored is not even a planet; it is a few grains of sand."
A Chronology of Computational Evolution
Verseon’s journey began in 2002, a time when the computational power required for atomic-level simulation was prohibitively expensive and rare. While contemporaries like Recursion and Exscientia later rode the wave of deep learning to public prominence, Verseon spent two decades building its "Deep Quantum Modeling" platform.
- 2002: Verseon is founded with a focus on physics-based computational design.
- 2015: The release of ZINC 15 highlights the explosion of purchasable chemical space.
- 2018: Verseon receives regulatory clearance for its lead anticoagulant candidate, VE-1902, to enter Phase I human trials.
- 2021: The company nominates VE-4840 as a primary candidate for diabetic retinopathy.
- 2025: Industry discourse shifts as meta-analyses show that many "AI-designed" molecules often lean on known chemical scaffolds, sparking a debate on the true novelty of machine-generated drugs.
The Physics-Based Paradigm vs. Generative AI
The core of Verseon’s philosophy is the distinction between "prediction" and "creation." Many modern AI platforms, including those popularized by protein-folding breakthroughs like AlphaFold, rely on interpolation. They perform exceptionally well because they are trained on dense, pre-existing experimental records.
Prakash is critical of the industry’s reliance on these models for de novo design. "AI is good at interpolation and terrible at extrapolation," he asserts. "An AI system by itself will not hand you something fundamentally new. If you train it on existing compounds and ask it to produce more, you get variations on what is already there."
To illustrate, he uses a automotive analogy: If medicinal chemistry is the car, many AI-discovered drugs are merely the same chassis with a new coat of paint or a different grille. This is why Verseon employs a different workflow. Their platform starts with the target—a protein pocket—and uses molecular and quantum-physics calculations to arrange atoms from the ground up to fit that pocket perfectly. Only after a novel structure is theoretically designed and validated through physics does the AI step in to assist in "tweaking" or generating variants for further optimization.
The Search for the "Holy Grail": Safe Anticoagulants
One of the most compelling applications of Verseon’s platform is its Precision Oral Anticoagulant (PROAC) program. For decades, the pharmaceutical industry has searched for a way to prevent dangerous blood clots without the side effect of major bleeding. While drugs like warfarin (introduced in the 1950s) and modern successors like Eliquis and Xarelto have saved lives, they still carry significant bleeding risks.
Verseon’s approach involves designing reversible covalent thrombin inhibitors that block clot formation while leaving the protein’s platelet-activating role untouched. By separating these functions, Verseon aims to provide an anticoagulant profile that mimics the body’s natural state, significantly reducing the risk of hemorrhage.
The company recently secured a new European patent for this program, underscoring the intellectual property value of their novel structural designs. Clinical data, including a 2020 study in Thrombosis Research, indicates that VE-1902 can achieve potent antithrombotic effects with a significantly improved safety profile compared to existing standards of care.
Diversifying the Pipeline: Beyond Clots to Vision
Verseon’s computational rigor is also being applied to ophthalmology. Diabetic macular edema, a complication of diabetes, often requires patients to receive frequent, invasive anti-VEGF injections directly into the eye.
Verseon is developing an oral alternative, VE-4840, which targets plasma kallikrein—a protein implicated in the fluid leakage that causes retinal damage. By moving from an injectable therapy to an oral, systemic one, Verseon hopes to change the standard of care for millions of patients, demonstrating that their platform is not restricted to a single therapeutic domain.
Implications for the Future of Drug Discovery
The debate over AI’s role in drug discovery reached a boiling point in 2022 when the Chemical Abstracts Service (CAS) analyzed the patent structures of early AI-designed clinical candidates. They found that a significant portion of these molecules shared structural shapes with existing, approved drugs—a finding that suggested a lack of true innovation.
As the industry matures, the "novelty standard" is becoming a critical metric. A 2025 review of 71 published cases found that ligand-based AI models frequently produced molecules with high structural similarity to known compounds. Verseon’s platform, by contrast, is designed to avoid this trap by ignoring "training data" as the starting point and instead relying on the laws of physics to generate chemistry that has never existed before.
While competitors like Insilico Medicine have made headlines by reaching Phase 3 trials with AI-identified targets, Verseon remains focused on the long game: the creation of proprietary, truly novel chemical matter.
Conclusion: The Path Ahead
Verseon stands at a crossroads. As the "AI bubble" in drug discovery faces the sobering reality of clinical trial outcomes, companies that have invested in the underlying physics of molecular interaction may find themselves with a distinct competitive advantage.
Whether Verseon’s specific candidates—VE-1902 and VE-4840—succeed in reaching the market remains to be seen. However, their insistence that drug discovery requires a synthesis of human scientific intuition, rigorous physics-based simulation, and controlled AI application provides a compelling roadmap. In an era where algorithms are often mistaken for wisdom, Verseon is betting that the most effective way to discover a new drug is not to ask a computer to guess what comes next, but to use science to build it from the atom up.
About the Author
