“An AlphaFold 4”: scientists marvel at the new AI exclusive to the drug spin-off DeepMind

Isomorphic Lab’s proprietary drug discovery model is a major breakthrough, but scientists developing open source tools wonder how to achieve similar results.

By Ewen Callaway & Nature magazine

Nearly two years after Google DeepMind released an update AlphaFold3 geared towards drug discoveryits biopharmaceutical spin-off, Isomorphic Labs, has announced an even more powerful artificial intelligence model – and they’re keeping it all to themselves.

London-based Isomorphic Labs touted the capabilities of its “drug discovery engine” – which it calls IsoDDE – in a 27-page document. technical report, published February 10. The achievements, including precise predictions of how proteins interact with potential drugs and antibody structures, have impressed scientists working in the field.

Yet unlike AlphaFold AI systems for predicting protein structure — which have been made available to other researchers and described in detail in journal articles — IsoDDE is proprietary, and the technical paper offers little insight into how to achieve similar results.

On supporting science journalism

If you enjoy this article, please consider supporting our award-winning journalism by subscribe. By purchasing a subscription, you are helping to ensure the future of impactful stories about the discoveries and ideas shaping our world today.

“This is a major breakthrough, on the scale of an AlphaFold4,” referring to a never-before-seen future generation of Google DeepMind technology, says Mohammed AlQuraishi, a computational biologist at Columbia University in New York who is working on developing fully open source versions of AlphaFold. “The problem, of course, is that we don’t know the details.”

Drug-protein interactions

AlphaFold 3 was developed with drug discovery in mind. Unlike his AlphaFold2, Nobel Prize-winning predecessorthe model could predict the structures of proteins interacting with other molecules, including potential drugs.

Similar AI modeled on AlphaFold 3 are almost at the height of its performance and have new capabilities. An open source model called Boltz-2, developed by scientists at the Massachusetts Institute of Technology in Cambridge and published last year, could predict the strength with which potential drugs bind to proteins, or binding affinity. This is a key property for therapeutic product development and is typically predicted by physics-based computationally intensive methods.

According to Isomorphic’s report, its new AI outperforms both Boltz-2 and physics-based methods in determining binding affinity. Predictions about how antibodies – which form the basis of therapies that generate tens of billions of pounds in sales each year – interact with their targets are also cutting edge, the report claims.

AlQuraishi says he is particularly impressed by IsoDDE’s ability to predict drug-protein interactions of molecules that are significantly different from the data the model was trained on. “That’s the most difficult problem and it suggests they had to do something quite new,” he says.

secret sauce

The models behind IsoDDE are “profoundly different” from other efforts, says Max Jaderberg, president of Isomorphic. But the company has no plans to reveal the “secret sauce” behind it all. “As with most big advances in machine learning and AI, it’s a combination of compute, data [and] algorithms,” Jaderberg adds. He hopes his team’s report will “galvanize” the efforts of other teams developing drug discovery AI.

“This report follows extensive efforts to collaborate with industry and potentially access their private structural data. So we do not know to what extent this additional data has an impact” on IsoDDE’s performance, Diego del Alamo, a computational structural biologist at Cambridge-based Takeda Pharmaceuticals, wrote on the social media site X.

Isomorphic has drug development deals, potentially worth billions of pounds, with pharmaceutical companies Johnson and Johnson, Eli Lilly and Novartis. It also has its own internal pipeline, with clinical trials on the horizon. Jaderberg says the company has developed different versions of IsoDDE from the one used for the technical report, including for work with its partners, which integrate different data sources.

His colleague Michael Schaarschmidt, director of machine learning at Isomorphic, says the company’s data strategy is “pretty comprehensive,” incorporating publicly available data, synthetic training data, and data sources that they will “try to license.”

Gabriele Corso, a machine learning scientist who co-developed Boltz-2 and who now runs the nonprofit Boltz in London, doesn’t think proprietary data played a critical role in the reported performance of Isomorphic’s tool, based on the gains his team saw. “There are many improvements we can make with the available data,” he says. “I think this is a new benchmark to match – but also to succeed.”

This article is reproduced with permission and has been published for the first time February 19, 2026.