DeepMind Spin-Off Unveils Proprietary 'AlphaFold 4'-Scale AI, Sparking Excitement and Secrecy Concerns in Drug Discovery

Global - Ekhbary News Agency

DeepMind Spin-Off Unveils Proprietary 'AlphaFold 4'-Scale AI, Sparking Excitement and Secrecy Concerns in Drug Discovery

In a move poised to reshape the landscape of pharmaceutical research, Isomorphic Labs, the London-based biopharmaceuticals spin-off of Google DeepMind, has announced a groundbreaking artificial intelligence model for drug discovery. Dubbed 'IsoDDE' (Isomorphic DeepMind Drug Engine), this proprietary AI is being hailed by some scientists as a generational leap, comparable to an "AlphaFold 4," due to its unprecedented capabilities in predicting complex molecular interactions. However, the company's decision to keep its underlying methodology under wraps has ignited a debate within the scientific community, which has long benefited from the open-source ethos of previous DeepMind projects.

The announcement, detailed in a 27-page technical report released on February 10, highlights IsoDDE's prowess in accurately forecasting how proteins interact with potential therapeutic molecules and the intricate structures of antibodies. These capabilities represent a significant advancement over existing tools, including DeepMind's own AlphaFold3, which was released nearly two years prior with a focus on drug discovery. AlphaFold3, unlike its Nobel Prize-winning predecessor AlphaFold2, was designed to predict the structures of proteins interacting with other molecules, including prospective drugs.

Mohammed AlQuraishi, a computational biologist at Columbia University in New York City, who is actively developing open-source versions of AlphaFold, expressed both awe and apprehension. "It’s a major advance, on the scale of an AlphaFold4," AlQuraishi noted, referring to an unreleased future generation of Google DeepMind’s technology. Yet, his enthusiasm is tempered by a critical caveat: "The problem, of course, is that we know nothing of the details." This sentiment echoes a broader concern among researchers who rely on transparency and open access to build upon foundational scientific discoveries.

The proprietary nature of IsoDDE stands in stark contrast to the AlphaFold AI systems for predicting protein structure, which were openly shared with the research community and thoroughly documented in peer-reviewed journals. While Isomorphic Labs' technical paper showcases impressive results, it offers minimal insight into the novel mechanisms that enable IsoDDE's superior performance, leaving independent scientists to speculate on how to replicate or even approach similar outcomes.

IsoDDE's report claims significant outperformance over both contemporary open-source models and traditional physics-based computational methods, particularly in determining binding affinity—the strength with which potential drugs attach to proteins. This is a crucial metric in drug development, typically requiring extensive computational resources. For instance, an open-source model called Boltz-2, developed by MIT scientists and released last year, also aimed to predict binding affinity and showed promising results. IsoDDE's reported superiority in this area, along with its state-of-the-art predictions for antibody-target interactions, underscores its potential to accelerate drug pipelines and generate substantial market value, given that antibody-based therapies alone account for tens of billions of pounds in annual sales.

Max Jaderberg, Isomorphic’s president, affirmed the distinctiveness of IsoDDE's underlying models, describing them as "profoundly different" from other efforts. However, he confirmed the company’s intention to keep its "secret sauce" proprietary, attributing its advancements to a synergy of "compute, data [and] algorithms." Jaderberg expressed hope that the report would nonetheless "galvanize" other teams engaged in drug-discovery AI development, suggesting a belief that the demonstrated capabilities, even without full disclosure, could inspire further innovation.

The role of proprietary data in IsoDDE's exceptional performance remains a point of speculation. Diego del Alamo, a computational structural biologist at Takeda Pharmaceuticals, highlighted this uncertainty, noting that Isomorphic's report follows "extensive efforts to partner with industry and potentially access their private structural data." Conversely, Gabriele Corso, a machine-learning scientist and co-developer of Boltz-2, believes that proprietary data may not be the sole differentiator. Based on the advancements his team is achieving with publicly available data, Corso suggested, "There are a lot of improvements we can make with the data that are out there. I think this is a new baseline to match — but also to pass."

Isomorphic Labs has already forged lucrative drug-development partnerships with pharmaceutical giants such as Johnson & Johnson, Eli Lilly, and Novartis, potentially worth billions of pounds. The company is also advancing its own internal drug pipeline, with clinical trials on the horizon. Jaderberg revealed that various versions of IsoDDE exist, tailored for specific partner collaborations and incorporating diverse data sources. Michael Schaarschmidt, Isomorphic’s director of machine learning, elaborated on the company’s "quite comprehensive" data strategy, which integrates public, synthetic, and licensed data. This multi-faceted approach to data acquisition and model development positions Isomorphic Labs at the forefront of AI-driven drug discovery, even as it navigates the delicate balance between scientific advancement and commercial confidentiality.

Ekhbary News Agency