As generative AI rapidly evolves, one of its most promising frontiers is transforming the early stages of drug development. Dr Ruth Gross, VP of business development at Evogene, shares how foundation models are unlocking vast chemical space, accelerating lead identification, and helping pharmaceutical partners optimize molecules for challenging therapeutic areas.

In the race to bring new treatments to market, time is one of the most valuable currencies—and foundation models are changing the way that currency is spent.

“Foundation models are a game-changer for small molecule drug discovery,” said Gross. 

“They enable the generation of novel compounds by learning from vast, diverse molecular data.”

While machine learning in drug discovery is nothing new, the sheer scale of today’s computational capabilities is what sets this new generation of models apart. 

“Until recently, scientists were limited to training models on millions of data sets,” Gross said. “Now we can train foundation models on billions of data parameters. That’s a seismic shift.”

This leap in capacity allows for the rapid simulation of biochemical scenarios and the virtual screening of enormous libraries of potential compounds. 

“You can test far more molecular compositions in far less time, across the entire drug development spectrum—whether it’s hit screening, hit-to-lead, or lead optimization,” she added.

Eliminating inefficiencies in early R&D

Using GPT-based systems, such as Evogene’s in-house platform, helps eliminate many of the manual modeling and lab testing methods that have historically slowed drug discovery.

“These models drastically reduce the number of false starts,” Gross explained. “Instead of spending months or years trying and failing to find a viable compound through traditional trial-and-error, foundation models offer better predictions, better optimizations, and better insights into molecular design.”

For pharmaceutical companies working to address unmet medical needs, that speed and accuracy can be transformative. “It lays the groundwork for getting potential new drugs to patients faster—and at lower cost,” she said.

Exploring the vast chemical universe

One of the most powerful applications of this technology is in exploring what Gross refers to as the “untapped chemical universe.”

“It’s estimated that less than 0.1% of all potentially synthesizable small molecules have been explored to date,” she said. “That’s a massive, largely uncharted space. Foundation models are helping scientists navigate it more efficiently than ever.”

Evogene’s AI platform, which is trained on more than 40 billion data points, integrates generative AI into de novo design and multi-parameter optimization—two essential aspects of modern drug development. This allows the model to make high-quality predictions and suggest optimized molecular designs in record time.

“We don’t just throw a model at a problem and hope for the best,” Gross explained. “We work closely with partners to define the specific molecular properties they’re looking for—whether that’s high solubility, improved safety, or better bioavailability. Then we generate blueprints for molecules that meet those exact needs.”

A multi-parameter puzzle, solved simultaneously

Drug development often involves solving multiple challenges at once—each one a complex problem in itself. According to Gross, foundation models are particularly well suited to this.

“Traditionally, if you tried to tweak one parameter—say, increase solubility—it might mess up another parameter like potency or safety,” she said. “It was a bit like solving a Rubik’s cube. Fix one side, and another goes wrong.”

But Evogene’s technology can balance these conflicting criteria in a single design cycle. “We can optimize across all the parameters a partner cares about—simultaneously—rather than going back and forth with iterative testing,” she said.

To illustrate the model’s flexibility, Gross pointed to how it adapts to different therapeutic goals:

• Central nervous system disorders: “If you’re developing a drug for something like Alzheimer’s, the compound often needs to cross the blood-brain barrier. That requires very specific lipophilic properties—and sometimes even the ability to be actively transported across,” she said. 
• Infectious diseases: “Antibiotics need to selectively target essential pathways in pathogens without harming the host. They also need to maintain effective concentrations at the site of infection.” 
• Cancer therapies: “With precision oncology, it’s all about selectivity. You want your molecule to inhibit a specific mutation without affecting the normal protein. You also need it to accumulate in tumor tissues.” 

Gross noted that these are just a few of the many competing demands drug developers must juggle. “That’s where AI really shines—because it can consider all of these at once and propose the best-balanced molecule.”

Cloud collaboration to scale innovation

Earlier this year, Evogene entered into a partnership with Google Cloud to further scale its AI capabilities. According to Gross, this collaboration has significantly accelerated model training and data processing.

“With Google Cloud’s scalable infrastructure and GPU-powered compute, we can now train our foundation models in weeks rather than months,” she said. “That was simply not possible with on-premise systems.”

The impact has been immediate. “We’ve tripled the number of data projects we’re able to take on at once,” Gross added. “And because we can analyze molecules so much faster, we’re delivering insights to our partners in days or weeks—instead of waiting six months or more.”

The regulatory shift toward AI

Despite the excitement around AI in drug discovery, pharmaceutical adoption hasn’t always kept pace with innovation. Gross acknowledged that some companies remain hesitant, often due to concerns about data quality, model transparency, or how AI integrates with existing workflows.

“Pharma is a very large, very regulated industry,” she said. “Change takes time. But we’re seeing real progress.”

Gross noted that regulators are beginning to recognize the value of AI in early-stage R&D, which is helping shift attitudes across the industry. “There’s more trust now that these tools can support—not replace—scientific decision-making.”

She believes that external partnerships, like those Evogene fosters, are helping bridge the gap. “We provide the agility and technical depth that pharma companies need, without asking them to overhaul their entire infrastructure overnight.”

What’s next for AI in drug development?

For all its predictive power, AI still faces a fundamental constraint: the need for biological validation. “The so-called wet lab bottleneck is real,” Gross said. “You still need to test your molecule in real biological systems to confirm it behaves as expected.”

But even here, change is on the horizon. “In-silico validation techniques are improving fast,” she noted. “We’re heading toward a future where some aspects of wet lab testing might be replaced—or at least heavily accelerated—by AI-driven simulations.”

Until then, close collaboration between computational and experimental teams remains key. “Our strongest results come from partnerships where AI guides the early design, and wet lab teams validate the outcomes. Together, they’re redefining what’s possible in drug discovery.”