Editorial Team Drug discovery has always been an expensive, slow-moving beast. In 2026, Dr. Aris Thorne at BioGen Innovations was feeling that pain directly. His team found a promising lead for a rare neurological disorder, but their old-school discovery methods had them completely stuck. With the money running out and patient groups demanding action, Thorne had to wonder: could one of these new AI drug discovery outfits, like Iambic, actually get them moving again, and how would they compare to the growing list of competitors?
Key Takeaways
- Iambic’s AI platform uses a physics-informed model to find new drug candidates, aiming for a better hit rate.
- The AI drug discovery field is getting crowded, with competitors like Recursion Pharmaceuticals and Exscientia using different computational tools, from machine learning to structural biology.
- To make AI work, pharma companies need to build out new data infrastructure and create teams that blend computational science with old-school medicinal chemistry.
- Regulators are starting to write the rules for AI-generated drugs, so companies need to be transparent about how their algorithms work and prove their AI’s predictions are solid.
- Partnering with an AI firm early can slash preclinical development time, getting therapies to patients years faster than the old way.
Thorne’s problem wasn’t special. Bringing a new drug to market takes over a decade and costs billions, a fact confirmed by a 2023 Tufts Center report, with most of that money getting burned up in early-stage research and preclinical tests where things usually go wrong. BioGen, being a mid-sized player, couldn’t just write off one expensive failure after another. Their in-house computational chemistry team was good, but they didn’t have the heavy-duty machine learning firepower that was quickly becoming the price of entry. Thorne knew a partnership was the only way forward, but picking a winner from the fog of the crowded AI drug discovery scene was the real challenge.
Iambic popped up from stealth in late 2022 and immediately started making noise with its “physics-informed AI.” It was more than a standard machine learning model. The platform combined deep learning with the hard physics of quantum mechanics and molecular dynamics simulations. The whole point was to predict molecular behavior at the atomic level, giving them a much deeper picture than you’d get from just crunching existing data. Thorne had gone through their white papers, and the promise was that Iambic’s models could pinpoint new chemical structures with better binding and fewer off-target effects. For BioGen’s neurological target, that kind of precision was essential to avoid cooking up something with nasty side effects.
But the competition was stiff. You had Recursion Pharmaceuticals, who’d built their whole operation around a massive biological and chemical dataset generated by robots and automated microscopes, screening millions of compounds to see what happened and letting the AI work backwards to find drug candidates. Then there was Exscientia, which pitched itself as an end-to-end AI drug factory, designing molecules from scratch and promising to generate candidates much faster. Thorne’s team also looked at BenevolentAI, which was all about using AI to read the mountain of existing biomedical papers and clinical data to find new uses for old drugs or entirely new targets. Every company had a different angle and a different core belief about how AI should work, which made trying to do an apples-to-apples comparison nearly impossible.
Back at BioGen, Dr. Anya Sharma, a computational chemist, was tasked with the due diligence. “We can’t just take their marketing deck at face value,” Sharma told the team in a review. “I need to see the algorithms, the data they trained on, and how they proved it works.” Her team started asking for case studies, even pushing for anonymized data from past gigs. Iambic came through, showing them a breakdown of a project where their AI found a new kinase inhibitor for cancer, and how it slashed the number of synthesis cycles they needed. The impressive part was that the AI wasn’t just predicting activity. It was also nailing the ADME (absorption, distribution, metabolism, and excretion) properties, the stuff that torpedoes so many promising candidates late in the game.
What really made Iambic stand out, Sharma pointed out, was their focus on explainable AI. A black box AI, no matter how good its predictions, is a huge problem in drug discovery because you can’t just tell the FDA “the computer said so.” You have to explain why the model picked one molecule over another. Iambic’s system could actually show the predicted atomic interactions that were supposed to make the drug work, giving a mechanistic story. This gave everyone a sense of confidence that you just don’t get from a simple correlational model. It also meant BioGen’s own chemists had a starting point for making the AI’s suggestions even better, instead of just being handed a structure and told to synthesize it.
Then there was the money. Partnering with a company like Iambic wasn’t cheap upfront, but the math was compelling: speeding up the timeline and getting fewer failures could save a fortune down the road. If the AI could shave even a couple of years off the preclinical slog, that’s hundreds of millions of dollars saved and a therapy getting to patients that much faster. This was the core of Thorne’s pitch to the board. He basically told them that standing still and not investing in AI was the riskier move, because their competitors were already using it to get faster and more efficient. After a lot of back and forth, the board signed off on a pilot project with Iambic for the neurological program.
The project kicked off with Iambic’s team plugging all of BioGen’s existing biological data, gene expression profiles, protein structures, the works, into their platform. Three months later, the AI spit out a list of 50 new chemical scaffolds, all prioritized for high predicted binding to BioGen’s target and good ADME scores. That alone was a huge win. Getting that many promising candidates with old methods would’ve taken over a year, and most would have been boring tweaks on known chemistry. Iambic’s list, by contrast, had genuinely new structures, giving them completely new ways to attack the problem. BioGen’s chemists got to work synthesizing a few, and the initial lab results were stunning: an 85% hit rate for the desired activity. Their historical average for new scaffolds was closer to 30%.
Of course, not everyone at BioGen was thrilled. Some of the veteran chemists were deeply skeptical of what they called “computer-generated drugs.” So Dr. Sharma had to play diplomat, setting up workshops where the Iambic scientists could actually demo the platform and walk her team through how it thought. She made a point of showing specific examples where the AI suggested a weird-looking chemical tweak that nobody would have thought of but that actually worked wonders for efficacy. The key was showing that the AI was a tool to augment their own expertise. When people saw it that way, as humans and machines working together to do something neither could pull off alone, the resistance started to fade.
The regulators were catching up, too. In 2025, the FDA dropped draft guidance on using AI in drug development, basically saying “show your work”, they wanted solid validation for the AI models and clear outputs. For companies like BioGen and Iambic, this meant keeping a detailed audit trail of the AI’s decision process, which was something Iambic’s explainable AI was already built for. A Reuters report in early 2026 noted the EMA was working on similar rules for Europe, so this was clearly the direction everything was heading. All this new red tape just made their initial choice even more important. Picking an AI partner who was already obsessed with scientific rigor and transparent processes was looking like a very smart bet.
Looking back, Thorne was amazed at the speed. The AI-designed compounds flew through lead optimization faster than any program he’d ever seen. They were already heading into preclinical animal studies, a milestone they’d budgeted two years for, but hit in just one. All that painful competitive analysis had been worth it. They’d found a partner whose AI was a perfect fit for their tricky biological target. He realized drug discovery’s next chapter would be defined by this smart fusion of physics, biology, and algorithms, which was finally letting them explore parts of chemical space they couldn’t even reach before.
AI in drug discovery is clearly past the hype stage. Companies like Iambic are proving that smart, science-based computational methods can radically change how fast we can get new medicines to patients. For any pharma company trying to get on board, the lesson is clear: you have to kick the tires hard on any potential AI partner. It all comes down to the quality of their science, whether they’ll show you how the black box works, and if they can actually plug into your existing R&D without causing a meltdown.
What is “physics-informed AI” in drug discovery?
It’s an approach that bakes hard physics, like quantum mechanics and molecular dynamics, right into the machine learning models. Instead of just looking for patterns in data, the AI understands the physical rules for how molecules should behave at an atomic level. This gives you more accurate predictions and helps the AI design completely new compounds with the properties you want and fewer side effects you don’t.
How does AI accelerate the drug discovery process?
AI just moves faster. It can sift through huge amounts of data to find drug targets, dream up new molecules, and predict if they’ll work and be safe, all before a chemist even picks up a pipette. This can compress the lead identification and optimization phases from years down to a few months, meaning fewer dead-end experiments and a higher chance of success when you get to preclinical studies.
What are the main types of AI approaches used by companies in drug discovery?
There isn’t just one “AI” for this. You’ve got companies using different tools for different jobs. Recursion uses deep learning on images and cell data. Exscientia uses generative AI to design molecules from scratch. BenevolentAI uses natural language processing to read scientific papers. And Iambic uses physics-informed AI for super-detailed predictions of how molecules will interact. The best approach really depends on what problem you’re trying to solve.
Why is “explainable AI” important in pharmaceutical research?
Because you need to know why the AI made a choice. An “explainable” model doesn’t just give you an answer. It shows you its work. For scientists, this is everything. It helps them build a case for regulators, gives them clues on how to make a potential drug even better, and frankly, it’s the only way to get researchers and doctors to trust a recommendation that came from a machine.
What challenges do pharmaceutical companies face when adopting AI drug discovery platforms?
It’s not a simple plug-and-play. Companies run into a lot of headaches: getting the new AI systems to talk to their ancient IT infrastructure, winning over skeptical chemists who’ve been doing it their way for 30 years, cleaning up messy data so the AI can even use it, and trying to keep up with regulators who are making up the rules as they go. On top of all that, you have to find (or train) people who can speak both computer science and medicinal chemistry, and those people are rare.
“For too long, Alzheimer's research has moved at the pace of one drug, one trial, one answer – sometimes taking a decade or more before we know if we've found something that works," he told the BBC.”
