Isomorphic Labs, the biopharmaceutical spin-off of Google DeepMind, has announced a powerful new AI model called IsoDDE that outperforms existing drug-discovery tools, but the company is keeping the technical details completely private, frustrating scientists trying to build competing open-source systems. The 27-page technical report released on February 10 describes achievements that have impressed researchers in the field, yet offers little insight into how the breakthrough was achieved. What Makes IsoDDE Different From AlphaFold 3? Nearly two years after Google DeepMind released AlphaFold 3, an updated system designed specifically for drug discovery, Isomorphic Labs unveiled what experts are calling a generational leap forward. "It's a major advance, on the scale of an AlphaFold 4," says Mohammed AlQuraishi, a computational biologist at Columbia University who is developing fully open-source versions of AlphaFold. "The problem, of course, is that we know nothing of the details." AlphaFold 3 was groundbreaking because it could predict how proteins interact with other molecules, including potential drugs. IsoDDE takes this further by achieving state-of-the-art performance on several critical drug-discovery tasks that were previously difficult to solve computationally. The tool can predict binding affinity, which measures how strongly potential drugs attach to proteins, a key property for developing therapeutics. It also excels at predicting how antibodies interact with their targets, a capability that matters enormously since antibody-based therapies generate tens of billions of pounds in annual sales. What particularly impressed AlQuraishi was IsoDDE's ability to predict drug-protein interactions for molecules vastly different from the data the model was trained on. "That's the really hard problem, and suggests that they must've done something pretty novel," he explains. Why Is Isomorphic Keeping Its AI Secret? Unlike AlphaFold, which was made accessible to researchers and described in depth in peer-reviewed journal articles, IsoDDE is proprietary. Isomorphic has no plans to reveal what the company calls the "secret sauce" behind the breakthrough. Max Jaderberg, Isomorphic's president, acknowledged that the model's success comes from "a combination of compute, data and algorithms," but declined to elaborate. He expressed hope that the technical report would "galvanize" efforts by other teams building drug-discovery AI systems. The secrecy reflects a fundamental shift in how AI breakthroughs are being commercialized. Isomorphic has struck drug-development deals potentially worth billions of pounds with pharmaceutical giants Johnson and Johnson, Eli Lilly, and Novartis. The company also maintains its own internal pipeline with clinical trials on the horizon. Jaderberg noted that Isomorphic has developed different versions of IsoDDE for work with its partners, incorporating different data sources. How Are Competitors Responding to the Proprietary Model? - Open-Source Alternatives: Gabriele Corso, who co-developed Boltz-2, an open-source competitor, believes proprietary data may not be essential to matching Isomorphic's performance. "There are a lot of improvements we can make with the data that are out there," he says. "I think this is a new baseline to match, but also to pass." - Data Strategy Questions: Diego del Alamo, a computational structural biologist at Takeda Pharmaceuticals, noted that Isomorphic's report came after extensive efforts to partner with industry and access private structural data. "We don't know how impactful that extra data is" to IsoDDE's performance, he wrote on social media. - Performance Benchmarks: According to Isomorphic's report, IsoDDE outperforms both Boltz-2 and physics-based methods at determining binding affinity, a task that usually requires computationally intensive calculations. The competitive landscape reveals a tension in AI development. Open-source models like Boltz-2 have come remarkably close to matching AlphaFold 3's performance and have added new capabilities, such as predicting binding affinity. However, Isomorphic's proprietary approach appears to have achieved something qualitatively different. Michael Schaarschmidt, Isomorphic's director of machine learning, described the company's data strategy as "quite comprehensive," incorporating publicly available data, synthetic training data, and data sources they plan to license. What Does This Mean for Drug Discovery? The emergence of IsoDDE highlights a critical moment in AI-driven drug discovery. For decades, computational biology relied on physics-based simulations that were slow and expensive. AI models like AlphaFold revolutionized the field by predicting protein structures with remarkable accuracy. IsoDDE represents the next frontier, automating predictions about how drugs interact with their targets, a process that typically requires expensive laboratory experiments. The proprietary nature of IsoDDE creates an interesting paradox. While it may accelerate drug discovery for Isomorphic's pharmaceutical partners, it leaves the broader scientific community without access to the methods that could advance the field more rapidly. Open-source alternatives continue to improve, but the gap between cutting-edge proprietary tools and publicly available systems is widening. This divide may ultimately determine which companies succeed in the race to discover new drugs faster and cheaper than traditional methods allow. For now, scientists working on open-source drug-discovery AI are treating IsoDDE as both inspiration and challenge. The technical report, while sparse on details, demonstrates what's possible when machine learning, computational power, and proprietary data converge. Whether competitors can replicate these results using only public data remains an open question that will shape the future of AI in medicine.
