Despite decades of progress in pharmaceutical R&D, the earliest phase of drug discovery—where molecules are conceived and designed—remains costly, inefficient, and burdened by high failure rates. Most experimental compounds never reach clinical development, and those that do often take over a decade and billions of dollars to bring to market.
These bottlenecks are rooted in the limitations of traditional methodologies, which rely heavily on screening vast libraries of known compounds in hopes of finding a match. At the same time, alternative approaches are gaining attention. Generative AI can be used to propose new molecular candidates during early discovery, shifting from search-and-filter to true design and invention. Variational AI, through its proprietary platform Enki™, is helping lead this transition.
Addressing bottlenecks in early-stage research
Historically, drug discovery has been dominated by high-throughput screening, where researchers sift through extensive compound libraries experimentally or, more recently, virtually, using discriminative algorithms to prioritize hits. This process, while increasingly digitized, is still governed by what already exists. As a result, it often takes years to move from hit identification to lead optimization and pre-clinical candidates, with a steep attrition curve along the way.
Furthermore, trial-and-error experimentation and siloed data can contribute to longer timelines and higher development costs. With average R&D expenditures exceeding $2.6 billion per approved drug, and timelines stretching over a decade, the need for innovation at the design stage is clear.
Generative AI offers a different approach to this phase—not by speeding up what exists, but by expanding the range of explored possibilities.
The strategic edge of generative over discriminative models
While many pharmaceutical companies have embraced AI, the majority rely on discriminative models that classify, rank, or predict properties of pre-existing molecules. These models are excellent filters—but they cannot create.
Generative models, in contrast, are engineered to invent. By learning the fundamental rules of medicinal chemistry, these models can propose entirely new molecular structures aligned to specific therapeutic goals.
Variational AI’s Enki™platform (Enki) exemplifies this generative-first approach. Purpose-built for drug discovery, Enki™supports the exploration of novel small molecules aligned with a target product profile (TPP).
This shift from optimization to invention is not just a technical upgrade—it represents a strategic inflection point for biopharma companies.
Multi-objective design: the true power of generative AI
The strength of generative models lies in their ability to jointly optimize for multiple objectives: potency, selectivity, safety, and synthetic accessibility at the moment of design. This can reduce the reliance on the need to iterate blindly through chemical space.
Enki™’s generative framework allows for faster iteration of molecules that are structurally novel yet pharmacologically relevant. By aligning generation with TPP constraints from the outset, the platform can help shorten certain early-stage timelines.
This shift has broad implications, not only supporting the generation of higher-quality molecules in established areas like oncology, but also may improve economic feasibility to pursue discovery programs in previously overlooked diseases where return on investment was once a barrier. More broadly, it signals a departure from the rigid funnels of legacy drug development frameworks toward a more expansive, hypothesis-driven model of discovery.
Industry momentum: strategic collaborations in action
Skepticism around AI-generated compounds has long been a barrier. Variational AI’s recent collaborations provide a strong signal of confidence from the industry’s top players. In 2025, the company announced a strategic collaboration with Merck (known as MSD outside of the United States and Canada) —one of the most respected names in pharma—underscoring a growing trust in generative platforms to contribute to meaningful discovery pipelines.
In parallel, the company’s work with Rakovina Therapeutics has progressed from lead generation to lead optimization, building on promising results. Compounds generated by Enki™have shown potential advantages in specific brain-penetrance use cases—an area of intense research for oncology and CNS indications.
These developments, along with recognition from Life Sciences BC as Emerging Company of the Year 2025, highlight how Variational AI’s technology is being put to the test, passing in both scientific and commercial arenas.
New models for platform-driven pharma partnerships
As generative platforms like Enki™scale, they may influence how organizations structure early research and development of drug discovery. Pharma companies are increasingly exploring discovery-stage collaborations where value lies not in licensing assets, but in gaining long-term access to AI-native infrastructure.
This opens the door to new engagement models—ones that reward platform capacity, not just compound delivery. For biotech firms, it offers a path to sustained relevance. For pharma, it may offer practical advantages in speed, novelty, and target fit.
Looking ahead, generative AI could help rebalance the R&D equation. Not only unlocking discovery in underexplored indications and rare diseases, but also elevating innovation within the therapeutic areas that are already intensely studied. By enabling the generation of more differentiated and purpose-built molecules, platforms such as Enki™ support broader and more detailed exploration.
A shift from optimization to innovation
Drug discovery has traditionally been a reductive process—narrowing down, filtering out, and optimizing within established constraints. Generative AI turns that on its head. It is an expansive force, enabling the intentional invention of molecules built for purpose.
Variational AI is proving that with the right foundation, bio pharma companies can design smarter, iterate faster, and ultimately bring better treatments to patients.
Learn more about Variational AI at: https://variational.ai/
This article is for informational purposes only and does not substitute for professional medical advice. If you are seeking medical advice, diagnosis or treatment, please consult a medical professional or healthcare provider.
VentureBeat newsroom and editorial staff were not involved in the creation of this content.