AI Drug-Design Engine 50% Faster, But No FDA Approval Yet

On February 10, Isomorphic Labs, the Google DeepMind spinoff focused on computational drug design, released IsoDDE: its Drug Design Engine. This is not a model or an AlphaFold upgrade, but a unified in silico drug discovery system that runs protein structure prediction, ligand binding, affinity estimation, and pocket identification in concert, generating in seconds what used to take days of physics-based simulation.

The computational gains are substantial. On the hardest molecular prediction tasks—the "Runs N' Poses" benchmark designed to test generalization to unfamiliar proteins—IsoDDE achieves a 50% success rate compared to AlphaFold 3's roughly 23%. On antibody-antigen modeling, IsoDDE beats AlphaFold 3 by 2.3× and the open-source Boltz-2 by 19.8×. On binding affinity prediction, it achieves a Pearson correlation of 0.85, beating the physics-based gold standard FEP+ at 0.78. These improvements suggest the computational bottleneck in drug design may no longer be the binding question.

Isomorphic has signed partnerships with Eli Lilly, Novartis, and Johnson & Johnson worth a combined $4 billion（約6400億円）+ in potential value. The financial structure, however, reveals industry caution. Eli Lilly paid $45 million（約72億円） upfront against $1.7 billion（約2700億円） in milestones; Novartis paid $37.5 million（約60億円） upfront against $1.2 billion（約1900億円）—a 50:1 ratio between promised "biobucks" and actual wire transfers. This ratio is standard across AI drug discovery deals in 2025. Pharma is enthusiastic enough to sign but cautious enough to make nearly all economics contingent on clinical results that don't exist yet. Novartis expanded its partnership in February 2025, doubling the number of programs to six, targeting what it described as "particularly challenging" and previously undruggable targets, on the same financial terms. The J&J deal, announced January 2026, covers small molecules, antibodies, peptides, and molecular glues.

The critical gap lies in clinical efficacy. Jayatunga et al. (2024), in the first systematic analysis of AI-discovered drugs in clinical trials, showed that AI-discovered molecules achieve 80–90% success rates in Phase I trials, well above the historical 40–65% average. AI is good at designing molecules that are safe and have decent pharmacokinetic properties: they get absorbed, distributed, metabolized, and excreted appropriately. Phase I is mostly about safety, and AI passes it. Phase II, however, tests whether the drug actually treats the disease. Here the numbers are sobering: AI-discovered drugs show roughly 40% Phase II success rates, about the same as traditionally discovered drugs. AI has not yet demonstrated it can predict whether a molecule will work in a patient, only that it can predict whether a molecule will be tolerable. If both trends hold, end-to-end success rates could rise from the historical 5–10% to something like 9–18%, which would roughly double R&D productivity. In a trillion-dollar industry, that matters enormously, though McKinsey estimates generative AI could generate $60–110 billion annually in economic value for pharma and medical products—a far cry from the narrative that AI will "solve" drug discovery.

Isomorphic's competitive position is unusual. It leads on computational benchmarks but trails on clinical progress. Insilico Medicine has the most advanced clinical portfolio: its IPF drug ISM001-055 (now called rentosertib) reached Phase IIa with positive results published in Nature Medicine in June 2025, and Insilico has 10+ investigational new drug approvals across 31 programs. Recursion Pharmaceuticals, which absorbed Exscientia in a $688 million（約1100億円） merger, runs millions of phenomics experiments weekly on 65 petabytes of biological imaging data and owns wet-lab infrastructure Isomorphic lacks. What Isomorphic has is the AlphaFold lineage, Alphabet-scale compute, and a unified architecture where each prediction task informs the others. On talent, it is staffing for clinical reality: the company hired Dr. Ben Wolf as Chief Medical Officer in June 2025, formerly at Relay Therapeutics with FDA approval experience for Ayvakit and Gavreto. It opened a Cambridge, Massachusetts office. The open-source threat is real but manageable in the near term. Chai Discovery (backed by OpenAI at a $1.3 billion（約2100億円） valuation, now partnered with Lilly on biologics) and Boltz (partnered with Pfizer) are making progress, but the gap is wide enough that Isomorphic has time, perhaps 18–24 months, to convert its computational lead into clinical evidence.

For Alphabet, Isomorphic is a rounding error that could become a franchise. The Other Bets segment posted a $3.6 billion（約5800億円） operating loss in 2025 against Alphabet's net income of $132 billion（約21兆円）. A $600 million（約960億円） funding round led by Thrive Capital in March 2025 suggests Alphabet understands the urgency of reaching the clinic, but Alphabet can sustain this bet indefinitely while the underlying science matures. Demis Hassabis cautioned: "We know we're never going to solve drug design with AlphaFold alone. We'll need half a dozen more breakthroughs of that magnitude." IsoDDE might be one of those breakthroughs. Isomorphic's first clinical candidates are targeted for late 2026. The clinical data, when it arrives, will tell us whether it is the kind of breakthrough that matters.

Isomorphic Labs occupies an unusual competitive position: it leads on computational benchmarks—beating AlphaFold 3 by 2.3× on antibody-antigen modeling and the open-source Boltz-2 by 19.8×—but trails rivals on clinical progress. Insilico Medicine has moved furthest clinically, with ISM001-055 (rentosertib) reaching Phase IIa with positive results published in Nature Medicine in June 2025 and 10+ investigational new drug approvals across 31 programs. Recursion Pharmaceuticals, which absorbed Exscientia in a $688 million（約1100億円） merger, pursues a different strategy: running millions of phenomics experiments weekly on 65 petabytes of biological imaging data. Both companies own wet-lab infrastructure that Isomorphic currently lacks.

The financial structure of pharma partnerships reveals cautious skepticism beneath the headlines. Eli Lilly paid $45 million（約72億円） upfront against $1.7 billion（約2700億円） in milestones, and Novartis paid $37.5 million（約60億円） upfront against $1.2 billion（約1900億円）—a 50:1 ratio of promised to actual cash. Novartis's February 2025 expansion, doubling programs to six on the same financial terms, is a positive signal that internal results impressed Novartis scientists, but such expansion and approved drugs remain separated by what the article calls "the most unforgiving filter in business: human biology." If both AI Phase I success (80–90%) and Phase II equivalence (40%) trends hold, end-to-end success rates could rise from the historical 5–10% to roughly 9–18%, which would roughly double R&D productivity. McKinsey estimates generative AI could generate $60–110 billion annually in economic value for pharma and medical products—a meaningful but far smaller claim than the narrative that generative AI will "solve" drug discovery.

The core technical question hinges on whether IsoDDE's ability to model induced fits (where proteins reshape to accommodate a drug) and identify cryptic binding pockets translates to better efficacy prediction. Demis Hassabis acknowledged that "we're never going to solve drug design with AlphaFold alone. We'll need half a dozen more breakthroughs of that magnitude." IsoDDE may be one such breakthrough, but the article notes this is plausible but unproven. Alphabet's position is asymmetric: Isomorphic is a rounding error against Alphabet's $132 billion（約21兆円） net income in 2025, allowing indefinite patience while the science matures—a luxury most biotech startups cannot afford. A $600 million（約960億円） funding round led by Thrive Capital in March 2025 signals urgency to reach the clinic, and hiring decisions (Dr. Ben Wolf as CMO in June 2025, opening a Cambridge office) show staffing for clinical reality rather than publication strategy.