Isomorphic Labs Bets on AI, Clinical Trials Remain

Isomorphic Labs aims to design drugs using AI and is preparing for its first preclinical steps. However, the development remains subject to a 90% failure rate and expert reservations. While the DeepMind subsidiary refines its tools, OpenAI and Anthropic are also making strides in drug discovery.

Frequent Failures and Scientific Reservations Persist

No AI-designed drug has received FDA approval or completed clinical trials, and about 90% of candidates fail. Setbacks are already affecting specialized players, with highly publicized molecules remaining in Phase 1. Voices are calling for caution: Emilia Javorsky reminds us that AI excels in rule-based or data-rich tasks, unlike biology, and criticizes overly rapid promises. For Lennard Lee, AI helps predict molecular properties and design, but the "ultimate test" remains the confrontation with biological reality. Charlotte Deane highlights the difficulty of modeling proteins with fluid structures and the lengthy process required to prove a treatment's effectiveness.

OpenAI and Anthropic Advance Their Tools and Partnerships

OpenAI introduced GPT-Rosalind in April to reduce the time between discovery and approval, while forming partnerships with Novo Nordisk and Moderna, and acquiring Orchestra Bio in October. Anthropic is preparing for an IPO, claims six partnerships in life sciences to deploy AI throughout the development cycle and accelerate trials, and aims for human trials by the end of 2028 for its own programs, including rare forms of pediatric muscular dystrophy. Dario Amodei believes that AI could cure most major diseases within five to ten years, claims to have survived early-stage cancer, and asserts that only concrete progress will address skepticism and the costs of AI.

Isomorphic Labs Aims to Predict Molecular Interactions at Scale

Isomorphic plans to leverage its models and computing capabilities to anticipate how a molecule interacts with all other molecules in the body, a scenario deemed too costly to test manually. In February, the company unveiled a design engine presented as an improvement over AlphaFold, intending to calibrate molecules to administer the minimum dose and avoid complications related to high blood concentrations. The stated goal is to achieve a level of precision such that the design phase no longer requires laboratory input. Internal programs target breakthrough treatments in oncology, for immune and inflammatory diseases. The success of AlphaFold is presented as an early advantage for the company.

Partnerships, Preclinical Work, and a Team in Boston, but a Cautious Timeline

Isomorphic has established collaborations with Novartis, Johnson & Johnson, and Eli Lilly. According to Max Jaderberg, the design engine now provides a foundation for preclinical testing, and human trials are "not so far away." However, he did not provide a timeline, which has already been pushed back, while indicating that no new delays have been recorded and that a clinical development team has been built in Boston over the past year in preparation for upcoming steps.

Bold Promises Amid an Intrinsically Long Journey

The development of a drug involves cycles of trial and error, frequent returns to the drawing board, few candidates tested in animals, and, after years, a minority making it to clinical trials. The process can exceed a decade, and even seasoned researchers often bring only one or two drugs to patients, while many do not succeed. Jaderberg attributes part of the failure to funding constraints pushing biotech companies toward imperfect molecules and off-target effects, but primarily points to an incomplete understanding of biology. He calls for building a mechanistic biological model and claims to be working in that direction. Isomorphic, a spin-off from DeepMind launched by Demis Hassabis, claims the legacy of AlphaFold, presented as having earned a Nobel Prize, to support this ambition.

Jaderberg's Journey and His Guiding Vision

President of Isomorphic Labs since January, Max Jaderberg manages operations under the leadership of Demis Hassabis. A graduate of Oxford and a founding team member, he was recruited to DeepMind after Hassabis acquired his computer vision startup launched during his studies. He reports that internal breakthroughs have shifted from a monthly to a weekly pace over three years and asserts that in five years, all drugs will be designed by AI, while claiming an increased focus of top researchers on impact. The mission he articulates for Isomorphic aims at drugs capable of curing cancer and eradicating diseases. The company relies on a team of researchers based in London.